16cbd5570SChris Mason /* 26cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 36cbd5570SChris Mason * 46cbd5570SChris Mason * This program is free software; you can redistribute it and/or 56cbd5570SChris Mason * modify it under the terms of the GNU General Public 66cbd5570SChris Mason * License v2 as published by the Free Software Foundation. 76cbd5570SChris Mason * 86cbd5570SChris Mason * This program is distributed in the hope that it will be useful, 96cbd5570SChris Mason * but WITHOUT ANY WARRANTY; without even the implied warranty of 106cbd5570SChris Mason * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 116cbd5570SChris Mason * General Public License for more details. 126cbd5570SChris Mason * 136cbd5570SChris Mason * You should have received a copy of the GNU General Public 146cbd5570SChris Mason * License along with this program; if not, write to the 156cbd5570SChris Mason * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 166cbd5570SChris Mason * Boston, MA 021110-1307, USA. 176cbd5570SChris Mason */ 186cbd5570SChris Mason 19e20d96d6SChris Mason #include <linux/fs.h> 20d98237b3SChris Mason #include <linux/blkdev.h> 2187cbda5cSChris Mason #include <linux/scatterlist.h> 2222b0ebdaSChris Mason #include <linux/swap.h> 230f7d52f4SChris Mason #include <linux/radix-tree.h> 2435b7e476SChris Mason #include <linux/writeback.h> 25d397712bSChris Mason #include <linux/buffer_head.h> 26ce9adaa5SChris Mason #include <linux/workqueue.h> 27a74a4b97SChris Mason #include <linux/kthread.h> 28a74a4b97SChris Mason #include <linux/freezer.h> 29163e783eSDavid Woodhouse #include <linux/crc32c.h> 305a0e3ad6STejun Heo #include <linux/slab.h> 31784b4e29SChris Mason #include <linux/migrate.h> 324b4e25f2SChris Mason #include "compat.h" 33eb60ceacSChris Mason #include "ctree.h" 34eb60ceacSChris Mason #include "disk-io.h" 35e089f05cSChris Mason #include "transaction.h" 360f7d52f4SChris Mason #include "btrfs_inode.h" 370b86a832SChris Mason #include "volumes.h" 38db94535dSChris Mason #include "print-tree.h" 398b712842SChris Mason #include "async-thread.h" 40925baeddSChris Mason #include "locking.h" 41e02119d5SChris Mason #include "tree-log.h" 42fa9c0d79SChris Mason #include "free-space-cache.h" 43eb60ceacSChris Mason 44d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops; 458b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 464df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root); 47acce952bSliubo static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 48acce952bSliubo int read_only); 49acce952bSliubo static int btrfs_destroy_ordered_operations(struct btrfs_root *root); 50acce952bSliubo static int btrfs_destroy_ordered_extents(struct btrfs_root *root); 51acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 52acce952bSliubo struct btrfs_root *root); 53acce952bSliubo static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t); 54acce952bSliubo static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 55acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 56acce952bSliubo struct extent_io_tree *dirty_pages, 57acce952bSliubo int mark); 58acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 59acce952bSliubo struct extent_io_tree *pinned_extents); 60acce952bSliubo static int btrfs_cleanup_transaction(struct btrfs_root *root); 61ce9adaa5SChris Mason 62d352ac68SChris Mason /* 63d352ac68SChris Mason * end_io_wq structs are used to do processing in task context when an IO is 64d352ac68SChris Mason * complete. This is used during reads to verify checksums, and it is used 65d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 66d352ac68SChris Mason */ 67ce9adaa5SChris Mason struct end_io_wq { 68ce9adaa5SChris Mason struct bio *bio; 69ce9adaa5SChris Mason bio_end_io_t *end_io; 70ce9adaa5SChris Mason void *private; 71ce9adaa5SChris Mason struct btrfs_fs_info *info; 72ce9adaa5SChris Mason int error; 7322c59948SChris Mason int metadata; 74ce9adaa5SChris Mason struct list_head list; 758b712842SChris Mason struct btrfs_work work; 76ce9adaa5SChris Mason }; 770da5468fSChris Mason 78d352ac68SChris Mason /* 79d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 80d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 81d352ac68SChris Mason * just before they are sent down the IO stack. 82d352ac68SChris Mason */ 8344b8bd7eSChris Mason struct async_submit_bio { 8444b8bd7eSChris Mason struct inode *inode; 8544b8bd7eSChris Mason struct bio *bio; 8644b8bd7eSChris Mason struct list_head list; 874a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start; 884a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done; 8944b8bd7eSChris Mason int rw; 9044b8bd7eSChris Mason int mirror_num; 91c8b97818SChris Mason unsigned long bio_flags; 92eaf25d93SChris Mason /* 93eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 94eaf25d93SChris Mason * can't tell us where in the file the bio should go 95eaf25d93SChris Mason */ 96eaf25d93SChris Mason u64 bio_offset; 978b712842SChris Mason struct btrfs_work work; 9844b8bd7eSChris Mason }; 9944b8bd7eSChris Mason 1004008c04aSChris Mason /* These are used to set the lockdep class on the extent buffer locks. 1014008c04aSChris Mason * The class is set by the readpage_end_io_hook after the buffer has 1024008c04aSChris Mason * passed csum validation but before the pages are unlocked. 1034008c04aSChris Mason * 1044008c04aSChris Mason * The lockdep class is also set by btrfs_init_new_buffer on freshly 1054008c04aSChris Mason * allocated blocks. 1064008c04aSChris Mason * 1074008c04aSChris Mason * The class is based on the level in the tree block, which allows lockdep 1084008c04aSChris Mason * to know that lower nodes nest inside the locks of higher nodes. 1094008c04aSChris Mason * 1104008c04aSChris Mason * We also add a check to make sure the highest level of the tree is 1114008c04aSChris Mason * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this 1124008c04aSChris Mason * code needs update as well. 1134008c04aSChris Mason */ 1144008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1154008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1164008c04aSChris Mason # error 1174008c04aSChris Mason # endif 1184008c04aSChris Mason static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1]; 1194008c04aSChris Mason static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = { 1204008c04aSChris Mason /* leaf */ 1214008c04aSChris Mason "btrfs-extent-00", 1224008c04aSChris Mason "btrfs-extent-01", 1234008c04aSChris Mason "btrfs-extent-02", 1244008c04aSChris Mason "btrfs-extent-03", 1254008c04aSChris Mason "btrfs-extent-04", 1264008c04aSChris Mason "btrfs-extent-05", 1274008c04aSChris Mason "btrfs-extent-06", 1284008c04aSChris Mason "btrfs-extent-07", 1294008c04aSChris Mason /* highest possible level */ 1304008c04aSChris Mason "btrfs-extent-08", 1314008c04aSChris Mason }; 1324008c04aSChris Mason #endif 1334008c04aSChris Mason 134d352ac68SChris Mason /* 135d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 136d352ac68SChris Mason * that covers the entire device 137d352ac68SChris Mason */ 138b2950863SChristoph Hellwig static struct extent_map *btree_get_extent(struct inode *inode, 139b2950863SChristoph Hellwig struct page *page, size_t page_offset, u64 start, u64 len, 1405f39d397SChris Mason int create) 1415f39d397SChris Mason { 1425f39d397SChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1435f39d397SChris Mason struct extent_map *em; 1445f39d397SChris Mason int ret; 1455f39d397SChris Mason 146890871beSChris Mason read_lock(&em_tree->lock); 147d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 148a061fc8dSChris Mason if (em) { 149a061fc8dSChris Mason em->bdev = 150a061fc8dSChris Mason BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 151890871beSChris Mason read_unlock(&em_tree->lock); 1525f39d397SChris Mason goto out; 153a061fc8dSChris Mason } 154890871beSChris Mason read_unlock(&em_tree->lock); 1557b13b7b1SChris Mason 1565f39d397SChris Mason em = alloc_extent_map(GFP_NOFS); 1575f39d397SChris Mason if (!em) { 1585f39d397SChris Mason em = ERR_PTR(-ENOMEM); 1595f39d397SChris Mason goto out; 1605f39d397SChris Mason } 1615f39d397SChris Mason em->start = 0; 1620afbaf8cSChris Mason em->len = (u64)-1; 163c8b97818SChris Mason em->block_len = (u64)-1; 1645f39d397SChris Mason em->block_start = 0; 165a061fc8dSChris Mason em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 166d1310b2eSChris Mason 167890871beSChris Mason write_lock(&em_tree->lock); 1685f39d397SChris Mason ret = add_extent_mapping(em_tree, em); 1695f39d397SChris Mason if (ret == -EEXIST) { 1700afbaf8cSChris Mason u64 failed_start = em->start; 1710afbaf8cSChris Mason u64 failed_len = em->len; 1720afbaf8cSChris Mason 1735f39d397SChris Mason free_extent_map(em); 1747b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 1750afbaf8cSChris Mason if (em) { 1767b13b7b1SChris Mason ret = 0; 1770afbaf8cSChris Mason } else { 1780afbaf8cSChris Mason em = lookup_extent_mapping(em_tree, failed_start, 1790afbaf8cSChris Mason failed_len); 1807b13b7b1SChris Mason ret = -EIO; 1810afbaf8cSChris Mason } 1825f39d397SChris Mason } else if (ret) { 1837b13b7b1SChris Mason free_extent_map(em); 1847b13b7b1SChris Mason em = NULL; 1855f39d397SChris Mason } 186890871beSChris Mason write_unlock(&em_tree->lock); 1877b13b7b1SChris Mason 1887b13b7b1SChris Mason if (ret) 1897b13b7b1SChris Mason em = ERR_PTR(ret); 1905f39d397SChris Mason out: 1915f39d397SChris Mason return em; 1925f39d397SChris Mason } 1935f39d397SChris Mason 19419c00ddcSChris Mason u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len) 19519c00ddcSChris Mason { 196163e783eSDavid Woodhouse return crc32c(seed, data, len); 19719c00ddcSChris Mason } 19819c00ddcSChris Mason 19919c00ddcSChris Mason void btrfs_csum_final(u32 crc, char *result) 20019c00ddcSChris Mason { 20119c00ddcSChris Mason *(__le32 *)result = ~cpu_to_le32(crc); 20219c00ddcSChris Mason } 20319c00ddcSChris Mason 204d352ac68SChris Mason /* 205d352ac68SChris Mason * compute the csum for a btree block, and either verify it or write it 206d352ac68SChris Mason * into the csum field of the block. 207d352ac68SChris Mason */ 20819c00ddcSChris Mason static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf, 20919c00ddcSChris Mason int verify) 21019c00ddcSChris Mason { 211607d432dSJosef Bacik u16 csum_size = 212607d432dSJosef Bacik btrfs_super_csum_size(&root->fs_info->super_copy); 213607d432dSJosef Bacik char *result = NULL; 21419c00ddcSChris Mason unsigned long len; 21519c00ddcSChris Mason unsigned long cur_len; 21619c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 21719c00ddcSChris Mason char *map_token = NULL; 21819c00ddcSChris Mason char *kaddr; 21919c00ddcSChris Mason unsigned long map_start; 22019c00ddcSChris Mason unsigned long map_len; 22119c00ddcSChris Mason int err; 22219c00ddcSChris Mason u32 crc = ~(u32)0; 223607d432dSJosef Bacik unsigned long inline_result; 22419c00ddcSChris Mason 22519c00ddcSChris Mason len = buf->len - offset; 22619c00ddcSChris Mason while (len > 0) { 22719c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 22819c00ddcSChris Mason &map_token, &kaddr, 22919c00ddcSChris Mason &map_start, &map_len, KM_USER0); 230d397712bSChris Mason if (err) 23119c00ddcSChris Mason return 1; 23219c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 23319c00ddcSChris Mason crc = btrfs_csum_data(root, kaddr + offset - map_start, 23419c00ddcSChris Mason crc, cur_len); 23519c00ddcSChris Mason len -= cur_len; 23619c00ddcSChris Mason offset += cur_len; 23719c00ddcSChris Mason unmap_extent_buffer(buf, map_token, KM_USER0); 23819c00ddcSChris Mason } 239607d432dSJosef Bacik if (csum_size > sizeof(inline_result)) { 240607d432dSJosef Bacik result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 241607d432dSJosef Bacik if (!result) 242607d432dSJosef Bacik return 1; 243607d432dSJosef Bacik } else { 244607d432dSJosef Bacik result = (char *)&inline_result; 245607d432dSJosef Bacik } 246607d432dSJosef Bacik 24719c00ddcSChris Mason btrfs_csum_final(crc, result); 24819c00ddcSChris Mason 24919c00ddcSChris Mason if (verify) { 250607d432dSJosef Bacik if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 251e4204dedSChris Mason u32 val; 252e4204dedSChris Mason u32 found = 0; 253607d432dSJosef Bacik memcpy(&found, result, csum_size); 254e4204dedSChris Mason 255607d432dSJosef Bacik read_extent_buffer(buf, &val, 0, csum_size); 256193f284dSChris Mason if (printk_ratelimit()) { 257193f284dSChris Mason printk(KERN_INFO "btrfs: %s checksum verify " 258193f284dSChris Mason "failed on %llu wanted %X found %X " 259193f284dSChris Mason "level %d\n", 26019c00ddcSChris Mason root->fs_info->sb->s_id, 261193f284dSChris Mason (unsigned long long)buf->start, val, found, 262193f284dSChris Mason btrfs_header_level(buf)); 263193f284dSChris Mason } 264607d432dSJosef Bacik if (result != (char *)&inline_result) 265607d432dSJosef Bacik kfree(result); 26619c00ddcSChris Mason return 1; 26719c00ddcSChris Mason } 26819c00ddcSChris Mason } else { 269607d432dSJosef Bacik write_extent_buffer(buf, result, 0, csum_size); 27019c00ddcSChris Mason } 271607d432dSJosef Bacik if (result != (char *)&inline_result) 272607d432dSJosef Bacik kfree(result); 27319c00ddcSChris Mason return 0; 27419c00ddcSChris Mason } 27519c00ddcSChris Mason 276d352ac68SChris Mason /* 277d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 278d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 279d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 280d352ac68SChris Mason * in the wrong place. 281d352ac68SChris Mason */ 2821259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 2831259ab75SChris Mason struct extent_buffer *eb, u64 parent_transid) 2841259ab75SChris Mason { 2852ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 2861259ab75SChris Mason int ret; 2871259ab75SChris Mason 2881259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 2891259ab75SChris Mason return 0; 2901259ab75SChris Mason 2912ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 2922ac55d41SJosef Bacik 0, &cached_state, GFP_NOFS); 2932ac55d41SJosef Bacik if (extent_buffer_uptodate(io_tree, eb, cached_state) && 2941259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 2951259ab75SChris Mason ret = 0; 2961259ab75SChris Mason goto out; 2971259ab75SChris Mason } 298193f284dSChris Mason if (printk_ratelimit()) { 299193f284dSChris Mason printk("parent transid verify failed on %llu wanted %llu " 300193f284dSChris Mason "found %llu\n", 3011259ab75SChris Mason (unsigned long long)eb->start, 3021259ab75SChris Mason (unsigned long long)parent_transid, 3031259ab75SChris Mason (unsigned long long)btrfs_header_generation(eb)); 304193f284dSChris Mason } 3051259ab75SChris Mason ret = 1; 3062ac55d41SJosef Bacik clear_extent_buffer_uptodate(io_tree, eb, &cached_state); 30733958dc6SChris Mason out: 3082ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 3092ac55d41SJosef Bacik &cached_state, GFP_NOFS); 3101259ab75SChris Mason return ret; 3111259ab75SChris Mason } 3121259ab75SChris Mason 313d352ac68SChris Mason /* 314d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 315d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 316d352ac68SChris Mason */ 317f188591eSChris Mason static int btree_read_extent_buffer_pages(struct btrfs_root *root, 318f188591eSChris Mason struct extent_buffer *eb, 319ca7a79adSChris Mason u64 start, u64 parent_transid) 320f188591eSChris Mason { 321f188591eSChris Mason struct extent_io_tree *io_tree; 322f188591eSChris Mason int ret; 323f188591eSChris Mason int num_copies = 0; 324f188591eSChris Mason int mirror_num = 0; 325f188591eSChris Mason 326a826d6dcSJosef Bacik clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 327f188591eSChris Mason io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 328f188591eSChris Mason while (1) { 329f188591eSChris Mason ret = read_extent_buffer_pages(io_tree, eb, start, 1, 330f188591eSChris Mason btree_get_extent, mirror_num); 3311259ab75SChris Mason if (!ret && 3321259ab75SChris Mason !verify_parent_transid(io_tree, eb, parent_transid)) 333f188591eSChris Mason return ret; 334d397712bSChris Mason 335a826d6dcSJosef Bacik /* 336a826d6dcSJosef Bacik * This buffer's crc is fine, but its contents are corrupted, so 337a826d6dcSJosef Bacik * there is no reason to read the other copies, they won't be 338a826d6dcSJosef Bacik * any less wrong. 339a826d6dcSJosef Bacik */ 340a826d6dcSJosef Bacik if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 341a826d6dcSJosef Bacik return ret; 342a826d6dcSJosef Bacik 343f188591eSChris Mason num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, 344f188591eSChris Mason eb->start, eb->len); 3454235298eSChris Mason if (num_copies == 1) 346f188591eSChris Mason return ret; 3474235298eSChris Mason 348f188591eSChris Mason mirror_num++; 3494235298eSChris Mason if (mirror_num > num_copies) 350f188591eSChris Mason return ret; 351f188591eSChris Mason } 352f188591eSChris Mason return -EIO; 353f188591eSChris Mason } 35419c00ddcSChris Mason 355d352ac68SChris Mason /* 356d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 357d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 358d352ac68SChris Mason */ 359d397712bSChris Mason 360b2950863SChristoph Hellwig static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 36119c00ddcSChris Mason { 362d1310b2eSChris Mason struct extent_io_tree *tree; 36335ebb934SChris Mason u64 start = (u64)page->index << PAGE_CACHE_SHIFT; 36419c00ddcSChris Mason u64 found_start; 36519c00ddcSChris Mason unsigned long len; 36619c00ddcSChris Mason struct extent_buffer *eb; 367f188591eSChris Mason int ret; 368f188591eSChris Mason 369d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 37019c00ddcSChris Mason 371eb14ab8eSChris Mason if (page->private == EXTENT_PAGE_PRIVATE) { 372eb14ab8eSChris Mason WARN_ON(1); 37319c00ddcSChris Mason goto out; 374eb14ab8eSChris Mason } 375eb14ab8eSChris Mason if (!page->private) { 376eb14ab8eSChris Mason WARN_ON(1); 37719c00ddcSChris Mason goto out; 378eb14ab8eSChris Mason } 37919c00ddcSChris Mason len = page->private >> 2; 380d397712bSChris Mason WARN_ON(len == 0); 381d397712bSChris Mason 38219c00ddcSChris Mason eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS); 38391ca338dSTsutomu Itoh if (eb == NULL) { 38491ca338dSTsutomu Itoh WARN_ON(1); 38591ca338dSTsutomu Itoh goto out; 38691ca338dSTsutomu Itoh } 387ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE, 388ca7a79adSChris Mason btrfs_header_generation(eb)); 389f188591eSChris Mason BUG_ON(ret); 390784b4e29SChris Mason WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN)); 391784b4e29SChris Mason 39219c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 39319c00ddcSChris Mason if (found_start != start) { 39455c69072SChris Mason WARN_ON(1); 39555c69072SChris Mason goto err; 39655c69072SChris Mason } 39755c69072SChris Mason if (eb->first_page != page) { 39855c69072SChris Mason WARN_ON(1); 39955c69072SChris Mason goto err; 40055c69072SChris Mason } 40155c69072SChris Mason if (!PageUptodate(page)) { 40255c69072SChris Mason WARN_ON(1); 40355c69072SChris Mason goto err; 40419c00ddcSChris Mason } 40519c00ddcSChris Mason csum_tree_block(root, eb, 0); 40655c69072SChris Mason err: 40719c00ddcSChris Mason free_extent_buffer(eb); 40819c00ddcSChris Mason out: 40919c00ddcSChris Mason return 0; 41019c00ddcSChris Mason } 41119c00ddcSChris Mason 4122b82032cSYan Zheng static int check_tree_block_fsid(struct btrfs_root *root, 4132b82032cSYan Zheng struct extent_buffer *eb) 4142b82032cSYan Zheng { 4152b82032cSYan Zheng struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 4162b82032cSYan Zheng u8 fsid[BTRFS_UUID_SIZE]; 4172b82032cSYan Zheng int ret = 1; 4182b82032cSYan Zheng 4192b82032cSYan Zheng read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb), 4202b82032cSYan Zheng BTRFS_FSID_SIZE); 4212b82032cSYan Zheng while (fs_devices) { 4222b82032cSYan Zheng if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 4232b82032cSYan Zheng ret = 0; 4242b82032cSYan Zheng break; 4252b82032cSYan Zheng } 4262b82032cSYan Zheng fs_devices = fs_devices->seed; 4272b82032cSYan Zheng } 4282b82032cSYan Zheng return ret; 4292b82032cSYan Zheng } 4302b82032cSYan Zheng 431a826d6dcSJosef Bacik #define CORRUPT(reason, eb, root, slot) \ 432a826d6dcSJosef Bacik printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \ 433a826d6dcSJosef Bacik "root=%llu, slot=%d\n", reason, \ 434a826d6dcSJosef Bacik (unsigned long long)btrfs_header_bytenr(eb), \ 435a826d6dcSJosef Bacik (unsigned long long)root->objectid, slot) 436a826d6dcSJosef Bacik 437a826d6dcSJosef Bacik static noinline int check_leaf(struct btrfs_root *root, 438a826d6dcSJosef Bacik struct extent_buffer *leaf) 439a826d6dcSJosef Bacik { 440a826d6dcSJosef Bacik struct btrfs_key key; 441a826d6dcSJosef Bacik struct btrfs_key leaf_key; 442a826d6dcSJosef Bacik u32 nritems = btrfs_header_nritems(leaf); 443a826d6dcSJosef Bacik int slot; 444a826d6dcSJosef Bacik 445a826d6dcSJosef Bacik if (nritems == 0) 446a826d6dcSJosef Bacik return 0; 447a826d6dcSJosef Bacik 448a826d6dcSJosef Bacik /* Check the 0 item */ 449a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 450a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 451a826d6dcSJosef Bacik CORRUPT("invalid item offset size pair", leaf, root, 0); 452a826d6dcSJosef Bacik return -EIO; 453a826d6dcSJosef Bacik } 454a826d6dcSJosef Bacik 455a826d6dcSJosef Bacik /* 456a826d6dcSJosef Bacik * Check to make sure each items keys are in the correct order and their 457a826d6dcSJosef Bacik * offsets make sense. We only have to loop through nritems-1 because 458a826d6dcSJosef Bacik * we check the current slot against the next slot, which verifies the 459a826d6dcSJosef Bacik * next slot's offset+size makes sense and that the current's slot 460a826d6dcSJosef Bacik * offset is correct. 461a826d6dcSJosef Bacik */ 462a826d6dcSJosef Bacik for (slot = 0; slot < nritems - 1; slot++) { 463a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 464a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot + 1); 465a826d6dcSJosef Bacik 466a826d6dcSJosef Bacik /* Make sure the keys are in the right order */ 467a826d6dcSJosef Bacik if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 468a826d6dcSJosef Bacik CORRUPT("bad key order", leaf, root, slot); 469a826d6dcSJosef Bacik return -EIO; 470a826d6dcSJosef Bacik } 471a826d6dcSJosef Bacik 472a826d6dcSJosef Bacik /* 473a826d6dcSJosef Bacik * Make sure the offset and ends are right, remember that the 474a826d6dcSJosef Bacik * item data starts at the end of the leaf and grows towards the 475a826d6dcSJosef Bacik * front. 476a826d6dcSJosef Bacik */ 477a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, slot) != 478a826d6dcSJosef Bacik btrfs_item_end_nr(leaf, slot + 1)) { 479a826d6dcSJosef Bacik CORRUPT("slot offset bad", leaf, root, slot); 480a826d6dcSJosef Bacik return -EIO; 481a826d6dcSJosef Bacik } 482a826d6dcSJosef Bacik 483a826d6dcSJosef Bacik /* 484a826d6dcSJosef Bacik * Check to make sure that we don't point outside of the leaf, 485a826d6dcSJosef Bacik * just incase all the items are consistent to eachother, but 486a826d6dcSJosef Bacik * all point outside of the leaf. 487a826d6dcSJosef Bacik */ 488a826d6dcSJosef Bacik if (btrfs_item_end_nr(leaf, slot) > 489a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 490a826d6dcSJosef Bacik CORRUPT("slot end outside of leaf", leaf, root, slot); 491a826d6dcSJosef Bacik return -EIO; 492a826d6dcSJosef Bacik } 493a826d6dcSJosef Bacik } 494a826d6dcSJosef Bacik 495a826d6dcSJosef Bacik return 0; 496a826d6dcSJosef Bacik } 497a826d6dcSJosef Bacik 4984008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 4994008c04aSChris Mason void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level) 5004008c04aSChris Mason { 5014008c04aSChris Mason lockdep_set_class_and_name(&eb->lock, 5024008c04aSChris Mason &btrfs_eb_class[level], 5034008c04aSChris Mason btrfs_eb_name[level]); 5044008c04aSChris Mason } 5054008c04aSChris Mason #endif 5064008c04aSChris Mason 507b2950863SChristoph Hellwig static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end, 508ce9adaa5SChris Mason struct extent_state *state) 509ce9adaa5SChris Mason { 510ce9adaa5SChris Mason struct extent_io_tree *tree; 511ce9adaa5SChris Mason u64 found_start; 512ce9adaa5SChris Mason int found_level; 513ce9adaa5SChris Mason unsigned long len; 514ce9adaa5SChris Mason struct extent_buffer *eb; 515ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 516f188591eSChris Mason int ret = 0; 517ce9adaa5SChris Mason 518ce9adaa5SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 519ce9adaa5SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) 520ce9adaa5SChris Mason goto out; 521ce9adaa5SChris Mason if (!page->private) 522ce9adaa5SChris Mason goto out; 523d397712bSChris Mason 524ce9adaa5SChris Mason len = page->private >> 2; 525d397712bSChris Mason WARN_ON(len == 0); 526d397712bSChris Mason 527ce9adaa5SChris Mason eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS); 52891ca338dSTsutomu Itoh if (eb == NULL) { 52991ca338dSTsutomu Itoh ret = -EIO; 53091ca338dSTsutomu Itoh goto out; 53191ca338dSTsutomu Itoh } 532f188591eSChris Mason 533ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 53423a07867SChris Mason if (found_start != start) { 535193f284dSChris Mason if (printk_ratelimit()) { 536193f284dSChris Mason printk(KERN_INFO "btrfs bad tree block start " 537193f284dSChris Mason "%llu %llu\n", 538a1b32a59SChris Mason (unsigned long long)found_start, 539a1b32a59SChris Mason (unsigned long long)eb->start); 540193f284dSChris Mason } 541f188591eSChris Mason ret = -EIO; 542ce9adaa5SChris Mason goto err; 543ce9adaa5SChris Mason } 544ce9adaa5SChris Mason if (eb->first_page != page) { 545d397712bSChris Mason printk(KERN_INFO "btrfs bad first page %lu %lu\n", 546d397712bSChris Mason eb->first_page->index, page->index); 547ce9adaa5SChris Mason WARN_ON(1); 548f188591eSChris Mason ret = -EIO; 549ce9adaa5SChris Mason goto err; 550ce9adaa5SChris Mason } 5512b82032cSYan Zheng if (check_tree_block_fsid(root, eb)) { 552193f284dSChris Mason if (printk_ratelimit()) { 553d397712bSChris Mason printk(KERN_INFO "btrfs bad fsid on block %llu\n", 554d397712bSChris Mason (unsigned long long)eb->start); 555193f284dSChris Mason } 5561259ab75SChris Mason ret = -EIO; 5571259ab75SChris Mason goto err; 5581259ab75SChris Mason } 559ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 560ce9adaa5SChris Mason 5614008c04aSChris Mason btrfs_set_buffer_lockdep_class(eb, found_level); 5624008c04aSChris Mason 563ce9adaa5SChris Mason ret = csum_tree_block(root, eb, 1); 564a826d6dcSJosef Bacik if (ret) { 565f188591eSChris Mason ret = -EIO; 566a826d6dcSJosef Bacik goto err; 567a826d6dcSJosef Bacik } 568a826d6dcSJosef Bacik 569a826d6dcSJosef Bacik /* 570a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 571a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 572a826d6dcSJosef Bacik * return -EIO. 573a826d6dcSJosef Bacik */ 574a826d6dcSJosef Bacik if (found_level == 0 && check_leaf(root, eb)) { 575a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 576a826d6dcSJosef Bacik ret = -EIO; 577a826d6dcSJosef Bacik } 578ce9adaa5SChris Mason 579ce9adaa5SChris Mason end = min_t(u64, eb->len, PAGE_CACHE_SIZE); 580ce9adaa5SChris Mason end = eb->start + end - 1; 581ce9adaa5SChris Mason err: 582ce9adaa5SChris Mason free_extent_buffer(eb); 583ce9adaa5SChris Mason out: 584f188591eSChris Mason return ret; 585ce9adaa5SChris Mason } 586ce9adaa5SChris Mason 587ce9adaa5SChris Mason static void end_workqueue_bio(struct bio *bio, int err) 588ce9adaa5SChris Mason { 589ce9adaa5SChris Mason struct end_io_wq *end_io_wq = bio->bi_private; 590ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 591ce9adaa5SChris Mason 592ce9adaa5SChris Mason fs_info = end_io_wq->info; 593ce9adaa5SChris Mason end_io_wq->error = err; 5948b712842SChris Mason end_io_wq->work.func = end_workqueue_fn; 5958b712842SChris Mason end_io_wq->work.flags = 0; 596d20f7043SChris Mason 5977b6d91daSChristoph Hellwig if (bio->bi_rw & REQ_WRITE) { 5980cb59c99SJosef Bacik if (end_io_wq->metadata == 1) 599cad321adSChris Mason btrfs_queue_worker(&fs_info->endio_meta_write_workers, 600cad321adSChris Mason &end_io_wq->work); 6010cb59c99SJosef Bacik else if (end_io_wq->metadata == 2) 6020cb59c99SJosef Bacik btrfs_queue_worker(&fs_info->endio_freespace_worker, 6030cb59c99SJosef Bacik &end_io_wq->work); 604cad321adSChris Mason else 605e6dcd2dcSChris Mason btrfs_queue_worker(&fs_info->endio_write_workers, 606e6dcd2dcSChris Mason &end_io_wq->work); 607d20f7043SChris Mason } else { 608d20f7043SChris Mason if (end_io_wq->metadata) 609d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 610d20f7043SChris Mason &end_io_wq->work); 611e6dcd2dcSChris Mason else 612d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_workers, 613d20f7043SChris Mason &end_io_wq->work); 614d20f7043SChris Mason } 615ce9adaa5SChris Mason } 616ce9adaa5SChris Mason 6170cb59c99SJosef Bacik /* 6180cb59c99SJosef Bacik * For the metadata arg you want 6190cb59c99SJosef Bacik * 6200cb59c99SJosef Bacik * 0 - if data 6210cb59c99SJosef Bacik * 1 - if normal metadta 6220cb59c99SJosef Bacik * 2 - if writing to the free space cache area 6230cb59c99SJosef Bacik */ 62422c59948SChris Mason int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 62522c59948SChris Mason int metadata) 6260b86a832SChris Mason { 627ce9adaa5SChris Mason struct end_io_wq *end_io_wq; 628ce9adaa5SChris Mason end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 629ce9adaa5SChris Mason if (!end_io_wq) 630ce9adaa5SChris Mason return -ENOMEM; 631ce9adaa5SChris Mason 632ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 633ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 63422c59948SChris Mason end_io_wq->info = info; 635ce9adaa5SChris Mason end_io_wq->error = 0; 636ce9adaa5SChris Mason end_io_wq->bio = bio; 63722c59948SChris Mason end_io_wq->metadata = metadata; 638ce9adaa5SChris Mason 639ce9adaa5SChris Mason bio->bi_private = end_io_wq; 640ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 64122c59948SChris Mason return 0; 64222c59948SChris Mason } 64322c59948SChris Mason 644b64a2851SChris Mason unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 6454854ddd0SChris Mason { 6464854ddd0SChris Mason unsigned long limit = min_t(unsigned long, 6474854ddd0SChris Mason info->workers.max_workers, 6484854ddd0SChris Mason info->fs_devices->open_devices); 6494854ddd0SChris Mason return 256 * limit; 6504854ddd0SChris Mason } 6514854ddd0SChris Mason 652777e6bd7SChris Mason int btrfs_congested_async(struct btrfs_fs_info *info, int iodone) 6530986fe9eSChris Mason { 654b64a2851SChris Mason return atomic_read(&info->nr_async_bios) > 655b64a2851SChris Mason btrfs_async_submit_limit(info); 6560986fe9eSChris Mason } 6570986fe9eSChris Mason 6584a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 6594a69a410SChris Mason { 6604a69a410SChris Mason struct async_submit_bio *async; 6614a69a410SChris Mason 6624a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6634a69a410SChris Mason async->submit_bio_start(async->inode, async->rw, async->bio, 664eaf25d93SChris Mason async->mirror_num, async->bio_flags, 665eaf25d93SChris Mason async->bio_offset); 6664a69a410SChris Mason } 6674a69a410SChris Mason 6684a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 6698b712842SChris Mason { 6708b712842SChris Mason struct btrfs_fs_info *fs_info; 6718b712842SChris Mason struct async_submit_bio *async; 6724854ddd0SChris Mason int limit; 6738b712842SChris Mason 6748b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 6758b712842SChris Mason fs_info = BTRFS_I(async->inode)->root->fs_info; 6764854ddd0SChris Mason 677b64a2851SChris Mason limit = btrfs_async_submit_limit(fs_info); 6784854ddd0SChris Mason limit = limit * 2 / 3; 6794854ddd0SChris Mason 6808b712842SChris Mason atomic_dec(&fs_info->nr_async_submits); 6810986fe9eSChris Mason 682b64a2851SChris Mason if (atomic_read(&fs_info->nr_async_submits) < limit && 683b64a2851SChris Mason waitqueue_active(&fs_info->async_submit_wait)) 6844854ddd0SChris Mason wake_up(&fs_info->async_submit_wait); 6854854ddd0SChris Mason 6864a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 687eaf25d93SChris Mason async->mirror_num, async->bio_flags, 688eaf25d93SChris Mason async->bio_offset); 6894a69a410SChris Mason } 6904a69a410SChris Mason 6914a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 6924a69a410SChris Mason { 6934a69a410SChris Mason struct async_submit_bio *async; 6944a69a410SChris Mason 6954a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 6968b712842SChris Mason kfree(async); 6978b712842SChris Mason } 6988b712842SChris Mason 69944b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 70044b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 701c8b97818SChris Mason unsigned long bio_flags, 702eaf25d93SChris Mason u64 bio_offset, 7034a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 7044a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 70544b8bd7eSChris Mason { 70644b8bd7eSChris Mason struct async_submit_bio *async; 70744b8bd7eSChris Mason 70844b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 70944b8bd7eSChris Mason if (!async) 71044b8bd7eSChris Mason return -ENOMEM; 71144b8bd7eSChris Mason 71244b8bd7eSChris Mason async->inode = inode; 71344b8bd7eSChris Mason async->rw = rw; 71444b8bd7eSChris Mason async->bio = bio; 71544b8bd7eSChris Mason async->mirror_num = mirror_num; 7164a69a410SChris Mason async->submit_bio_start = submit_bio_start; 7174a69a410SChris Mason async->submit_bio_done = submit_bio_done; 7184a69a410SChris Mason 7194a69a410SChris Mason async->work.func = run_one_async_start; 7204a69a410SChris Mason async->work.ordered_func = run_one_async_done; 7214a69a410SChris Mason async->work.ordered_free = run_one_async_free; 7224a69a410SChris Mason 7238b712842SChris Mason async->work.flags = 0; 724c8b97818SChris Mason async->bio_flags = bio_flags; 725eaf25d93SChris Mason async->bio_offset = bio_offset; 7268c8bee1dSChris Mason 727cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 728d313d7a3SChris Mason 7297b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 730d313d7a3SChris Mason btrfs_set_work_high_prio(&async->work); 731d313d7a3SChris Mason 7328b712842SChris Mason btrfs_queue_worker(&fs_info->workers, &async->work); 7339473f16cSChris Mason 734771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 735771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 736771ed689SChris Mason wait_event(fs_info->async_submit_wait, 737771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 738771ed689SChris Mason } 739771ed689SChris Mason 74044b8bd7eSChris Mason return 0; 74144b8bd7eSChris Mason } 74244b8bd7eSChris Mason 743ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 744ce3ed71aSChris Mason { 745ce3ed71aSChris Mason struct bio_vec *bvec = bio->bi_io_vec; 746ce3ed71aSChris Mason int bio_index = 0; 747ce3ed71aSChris Mason struct btrfs_root *root; 748ce3ed71aSChris Mason 749ce3ed71aSChris Mason WARN_ON(bio->bi_vcnt <= 0); 750ce3ed71aSChris Mason while (bio_index < bio->bi_vcnt) { 751ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 752ce3ed71aSChris Mason csum_dirty_buffer(root, bvec->bv_page); 753ce3ed71aSChris Mason bio_index++; 754ce3ed71aSChris Mason bvec++; 755ce3ed71aSChris Mason } 756ce3ed71aSChris Mason return 0; 757ce3ed71aSChris Mason } 758ce3ed71aSChris Mason 7594a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 7604a69a410SChris Mason struct bio *bio, int mirror_num, 761eaf25d93SChris Mason unsigned long bio_flags, 762eaf25d93SChris Mason u64 bio_offset) 76322c59948SChris Mason { 7648b712842SChris Mason /* 7658b712842SChris Mason * when we're called for a write, we're already in the async 7665443be45SChris Mason * submission context. Just jump into btrfs_map_bio 7678b712842SChris Mason */ 768ce3ed71aSChris Mason btree_csum_one_bio(bio); 7694a69a410SChris Mason return 0; 77022c59948SChris Mason } 77122c59948SChris Mason 7724a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 773eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 774eaf25d93SChris Mason u64 bio_offset) 7754a69a410SChris Mason { 7768b712842SChris Mason /* 7774a69a410SChris Mason * when we're called for a write, we're already in the async 7784a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 7798b712842SChris Mason */ 7808b712842SChris Mason return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 7810b86a832SChris Mason } 7820b86a832SChris Mason 78344b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 784eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 785eaf25d93SChris Mason u64 bio_offset) 78644b8bd7eSChris Mason { 7874a69a410SChris Mason int ret; 788cad321adSChris Mason 7894a69a410SChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 7904a69a410SChris Mason bio, 1); 7914a69a410SChris Mason BUG_ON(ret); 7924a69a410SChris Mason 7937b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 794cad321adSChris Mason /* 795cad321adSChris Mason * called for a read, do the setup so that checksum validation 796cad321adSChris Mason * can happen in the async kernel threads 797cad321adSChris Mason */ 7984a69a410SChris Mason return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 7996f3577bdSChris Mason mirror_num, 0); 80044b8bd7eSChris Mason } 801d313d7a3SChris Mason 802cad321adSChris Mason /* 803cad321adSChris Mason * kthread helpers are used to submit writes so that checksumming 804cad321adSChris Mason * can happen in parallel across all CPUs 805cad321adSChris Mason */ 80644b8bd7eSChris Mason return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 807c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 808eaf25d93SChris Mason bio_offset, 8094a69a410SChris Mason __btree_submit_bio_start, 8104a69a410SChris Mason __btree_submit_bio_done); 81144b8bd7eSChris Mason } 81244b8bd7eSChris Mason 8133dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 814784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 815784b4e29SChris Mason struct page *newpage, struct page *page) 816784b4e29SChris Mason { 817784b4e29SChris Mason /* 818784b4e29SChris Mason * we can't safely write a btree page from here, 819784b4e29SChris Mason * we haven't done the locking hook 820784b4e29SChris Mason */ 821784b4e29SChris Mason if (PageDirty(page)) 822784b4e29SChris Mason return -EAGAIN; 823784b4e29SChris Mason /* 824784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 825784b4e29SChris Mason * We must have no buffers or drop them. 826784b4e29SChris Mason */ 827784b4e29SChris Mason if (page_has_private(page) && 828784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 829784b4e29SChris Mason return -EAGAIN; 830784b4e29SChris Mason return migrate_page(mapping, newpage, page); 831784b4e29SChris Mason } 8323dd1462eSJan Beulich #endif 833784b4e29SChris Mason 8345f39d397SChris Mason static int btree_writepage(struct page *page, struct writeback_control *wbc) 8355f39d397SChris Mason { 836d1310b2eSChris Mason struct extent_io_tree *tree; 837b9473439SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 838b9473439SChris Mason struct extent_buffer *eb; 839b9473439SChris Mason int was_dirty; 8405443be45SChris Mason 841b9473439SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 842b9473439SChris Mason if (!(current->flags & PF_MEMALLOC)) { 843b9473439SChris Mason return extent_write_full_page(tree, page, 844b9473439SChris Mason btree_get_extent, wbc); 845b9473439SChris Mason } 846b9473439SChris Mason 8475443be45SChris Mason redirty_page_for_writepage(wbc, page); 848784b4e29SChris Mason eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE); 849b9473439SChris Mason WARN_ON(!eb); 850b9473439SChris Mason 851b9473439SChris Mason was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); 852b9473439SChris Mason if (!was_dirty) { 853b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 854b9473439SChris Mason root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE; 855b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 856b9473439SChris Mason } 857b9473439SChris Mason free_extent_buffer(eb); 858b9473439SChris Mason 8595443be45SChris Mason unlock_page(page); 8605443be45SChris Mason return 0; 8615443be45SChris Mason } 8620da5468fSChris Mason 8630da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 8640da5468fSChris Mason struct writeback_control *wbc) 8650da5468fSChris Mason { 866d1310b2eSChris Mason struct extent_io_tree *tree; 867d1310b2eSChris Mason tree = &BTRFS_I(mapping->host)->io_tree; 868d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 869b9473439SChris Mason struct btrfs_root *root = BTRFS_I(mapping->host)->root; 870793955bcSChris Mason u64 num_dirty; 87124ab9cd8SChris Mason unsigned long thresh = 32 * 1024 * 1024; 872448d640bSChris Mason 873448d640bSChris Mason if (wbc->for_kupdate) 874448d640bSChris Mason return 0; 875448d640bSChris Mason 876b9473439SChris Mason /* this is a bit racy, but that's ok */ 877b9473439SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 878d397712bSChris Mason if (num_dirty < thresh) 879793955bcSChris Mason return 0; 880793955bcSChris Mason } 8810da5468fSChris Mason return extent_writepages(tree, mapping, btree_get_extent, wbc); 8820da5468fSChris Mason } 8830da5468fSChris Mason 884b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 8855f39d397SChris Mason { 886d1310b2eSChris Mason struct extent_io_tree *tree; 887d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 8885f39d397SChris Mason return extent_read_full_page(tree, page, btree_get_extent); 8895f39d397SChris Mason } 8905f39d397SChris Mason 89170dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 8925f39d397SChris Mason { 893d1310b2eSChris Mason struct extent_io_tree *tree; 894d1310b2eSChris Mason struct extent_map_tree *map; 8955f39d397SChris Mason int ret; 8965f39d397SChris Mason 89798509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 89898509cfcSChris Mason return 0; 89998509cfcSChris Mason 900d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 901d1310b2eSChris Mason map = &BTRFS_I(page->mapping->host)->extent_tree; 9026af118ceSChris Mason 9037b13b7b1SChris Mason ret = try_release_extent_state(map, tree, page, gfp_flags); 904d397712bSChris Mason if (!ret) 9056af118ceSChris Mason return 0; 9066af118ceSChris Mason 9076af118ceSChris Mason ret = try_release_extent_buffer(tree, page); 9085f39d397SChris Mason if (ret == 1) { 9095f39d397SChris Mason ClearPagePrivate(page); 9105f39d397SChris Mason set_page_private(page, 0); 911d98237b3SChris Mason page_cache_release(page); 9125f39d397SChris Mason } 9136af118ceSChris Mason 914d98237b3SChris Mason return ret; 915d98237b3SChris Mason } 916d98237b3SChris Mason 9175f39d397SChris Mason static void btree_invalidatepage(struct page *page, unsigned long offset) 918d98237b3SChris Mason { 919d1310b2eSChris Mason struct extent_io_tree *tree; 920d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 9215f39d397SChris Mason extent_invalidatepage(tree, page, offset); 9225f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 9239ad6b7bcSChris Mason if (PagePrivate(page)) { 924d397712bSChris Mason printk(KERN_WARNING "btrfs warning page private not zero " 925d397712bSChris Mason "on page %llu\n", (unsigned long long)page_offset(page)); 9269ad6b7bcSChris Mason ClearPagePrivate(page); 9279ad6b7bcSChris Mason set_page_private(page, 0); 9289ad6b7bcSChris Mason page_cache_release(page); 9299ad6b7bcSChris Mason } 930d98237b3SChris Mason } 931d98237b3SChris Mason 9327f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 933d98237b3SChris Mason .readpage = btree_readpage, 934d98237b3SChris Mason .writepage = btree_writepage, 9350da5468fSChris Mason .writepages = btree_writepages, 9365f39d397SChris Mason .releasepage = btree_releasepage, 9375f39d397SChris Mason .invalidatepage = btree_invalidatepage, 938d98237b3SChris Mason .sync_page = block_sync_page, 9395a92bc88SChris Mason #ifdef CONFIG_MIGRATION 940784b4e29SChris Mason .migratepage = btree_migratepage, 9415a92bc88SChris Mason #endif 942d98237b3SChris Mason }; 943123abc88SChris Mason 944ca7a79adSChris Mason int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 945ca7a79adSChris Mason u64 parent_transid) 946090d1875SChris Mason { 9475f39d397SChris Mason struct extent_buffer *buf = NULL; 9485f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 949de428b63SChris Mason int ret = 0; 950090d1875SChris Mason 951db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 9525f39d397SChris Mason if (!buf) 953090d1875SChris Mason return 0; 954d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 955f188591eSChris Mason buf, 0, 0, btree_get_extent, 0); 9565f39d397SChris Mason free_extent_buffer(buf); 957de428b63SChris Mason return ret; 958090d1875SChris Mason } 959090d1875SChris Mason 9600999df54SChris Mason struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 9610999df54SChris Mason u64 bytenr, u32 blocksize) 9620999df54SChris Mason { 9630999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 9640999df54SChris Mason struct extent_buffer *eb; 9650999df54SChris Mason eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 9660999df54SChris Mason bytenr, blocksize, GFP_NOFS); 9670999df54SChris Mason return eb; 9680999df54SChris Mason } 9690999df54SChris Mason 9700999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 9710999df54SChris Mason u64 bytenr, u32 blocksize) 9720999df54SChris Mason { 9730999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 9740999df54SChris Mason struct extent_buffer *eb; 9750999df54SChris Mason 9760999df54SChris Mason eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 9770999df54SChris Mason bytenr, blocksize, NULL, GFP_NOFS); 9780999df54SChris Mason return eb; 9790999df54SChris Mason } 9800999df54SChris Mason 9810999df54SChris Mason 982e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 983e02119d5SChris Mason { 9848aa38c31SChristoph Hellwig return filemap_fdatawrite_range(buf->first_page->mapping, buf->start, 9858aa38c31SChristoph Hellwig buf->start + buf->len - 1); 986e02119d5SChris Mason } 987e02119d5SChris Mason 988e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 989e02119d5SChris Mason { 9908aa38c31SChristoph Hellwig return filemap_fdatawait_range(buf->first_page->mapping, 9918aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 992e02119d5SChris Mason } 993e02119d5SChris Mason 994db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 995ca7a79adSChris Mason u32 blocksize, u64 parent_transid) 996e20d96d6SChris Mason { 9975f39d397SChris Mason struct extent_buffer *buf = NULL; 99819c00ddcSChris Mason int ret; 99919c00ddcSChris Mason 1000db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 10015f39d397SChris Mason if (!buf) 1002d98237b3SChris Mason return NULL; 1003e4204dedSChris Mason 1004ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 1005ce9adaa5SChris Mason 1006d397712bSChris Mason if (ret == 0) 1007b4ce94deSChris Mason set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); 10085f39d397SChris Mason return buf; 1009ce9adaa5SChris Mason 1010eb60ceacSChris Mason } 1011eb60ceacSChris Mason 1012e089f05cSChris Mason int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 10135f39d397SChris Mason struct extent_buffer *buf) 1014ed2ff2cbSChris Mason { 10155f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 101655c69072SChris Mason if (btrfs_header_generation(buf) == 1017925baeddSChris Mason root->fs_info->running_transaction->transid) { 1018b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1019b4ce94deSChris Mason 1020b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1021b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 1022b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= buf->len) 1023b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= buf->len; 1024b9473439SChris Mason else 1025b9473439SChris Mason WARN_ON(1); 1026b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 1027b9473439SChris Mason } 1028b4ce94deSChris Mason 1029b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1030b9473439SChris Mason btrfs_set_lock_blocking(buf); 1031d1310b2eSChris Mason clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 103255c69072SChris Mason buf); 1033925baeddSChris Mason } 10345f39d397SChris Mason return 0; 10355f39d397SChris Mason } 10365f39d397SChris Mason 1037db94535dSChris Mason static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 103887ee04ebSChris Mason u32 stripesize, struct btrfs_root *root, 10399f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 1040e20d96d6SChris Mason u64 objectid) 1041d97e63b6SChris Mason { 1042cfaa7295SChris Mason root->node = NULL; 1043a28ec197SChris Mason root->commit_root = NULL; 1044db94535dSChris Mason root->sectorsize = sectorsize; 1045db94535dSChris Mason root->nodesize = nodesize; 1046db94535dSChris Mason root->leafsize = leafsize; 104787ee04ebSChris Mason root->stripesize = stripesize; 1048123abc88SChris Mason root->ref_cows = 0; 10490b86a832SChris Mason root->track_dirty = 0; 1050c71bf099SYan, Zheng root->in_radix = 0; 1051d68fc57bSYan, Zheng root->orphan_item_inserted = 0; 1052d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 10530b86a832SChris Mason 10549f5fae2fSChris Mason root->fs_info = fs_info; 10550f7d52f4SChris Mason root->objectid = objectid; 10560f7d52f4SChris Mason root->last_trans = 0; 105713a8a7c8SYan, Zheng root->highest_objectid = 0; 105858176a96SJosef Bacik root->name = NULL; 10594313b399SChris Mason root->in_sysfs = 0; 10606bef4d31SEric Paris root->inode_tree = RB_ROOT; 1061f0486c68SYan, Zheng root->block_rsv = NULL; 1062d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 10630b86a832SChris Mason 10640b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 10657b128766SJosef Bacik INIT_LIST_HEAD(&root->orphan_list); 10665d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1067925baeddSChris Mason spin_lock_init(&root->node_lock); 1068d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 10695d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1070f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 1071a2135011SChris Mason mutex_init(&root->objectid_mutex); 1072e02119d5SChris Mason mutex_init(&root->log_mutex); 10737237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 10747237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 10757237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 10767237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 10777237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 10787237f183SYan Zheng atomic_set(&root->log_writers, 0); 10797237f183SYan Zheng root->log_batch = 0; 10807237f183SYan Zheng root->log_transid = 0; 1081257c62e1SChris Mason root->last_log_commit = 0; 1082d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1083d0c803c4SChris Mason fs_info->btree_inode->i_mapping, GFP_NOFS); 1084017e5369SChris Mason 10853768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 10863768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 10876702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 108858176a96SJosef Bacik memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 10893f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 109058176a96SJosef Bacik init_completion(&root->kobj_unregister); 10916702ed49SChris Mason root->defrag_running = 0; 10924d775673SChris Mason root->root_key.objectid = objectid; 10933394e160SChris Mason root->anon_super.s_root = NULL; 10943394e160SChris Mason root->anon_super.s_dev = 0; 10953394e160SChris Mason INIT_LIST_HEAD(&root->anon_super.s_list); 10963394e160SChris Mason INIT_LIST_HEAD(&root->anon_super.s_instances); 10973394e160SChris Mason init_rwsem(&root->anon_super.s_umount); 10983394e160SChris Mason 10993768f368SChris Mason return 0; 11003768f368SChris Mason } 11013768f368SChris Mason 1102db94535dSChris Mason static int find_and_setup_root(struct btrfs_root *tree_root, 11039f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 11049f5fae2fSChris Mason u64 objectid, 1105e20d96d6SChris Mason struct btrfs_root *root) 11063768f368SChris Mason { 11073768f368SChris Mason int ret; 1108db94535dSChris Mason u32 blocksize; 110984234f3aSYan Zheng u64 generation; 11103768f368SChris Mason 1111db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 111287ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 111387ee04ebSChris Mason root, fs_info, objectid); 11143768f368SChris Mason ret = btrfs_find_last_root(tree_root, objectid, 11153768f368SChris Mason &root->root_item, &root->root_key); 11164df27c4dSYan, Zheng if (ret > 0) 11174df27c4dSYan, Zheng return -ENOENT; 11183768f368SChris Mason BUG_ON(ret); 11193768f368SChris Mason 112084234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1121db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1122db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 112384234f3aSYan Zheng blocksize, generation); 112468433b73SChris Mason if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) { 112568433b73SChris Mason free_extent_buffer(root->node); 112668433b73SChris Mason return -EIO; 112768433b73SChris Mason } 11284df27c4dSYan, Zheng root->commit_root = btrfs_root_node(root); 1129d97e63b6SChris Mason return 0; 1130d97e63b6SChris Mason } 1131d97e63b6SChris Mason 11327237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1133e02119d5SChris Mason struct btrfs_fs_info *fs_info) 11340f7d52f4SChris Mason { 11350f7d52f4SChris Mason struct btrfs_root *root; 11360f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 11377237f183SYan Zheng struct extent_buffer *leaf; 1138e02119d5SChris Mason 1139e02119d5SChris Mason root = kzalloc(sizeof(*root), GFP_NOFS); 1140e02119d5SChris Mason if (!root) 11417237f183SYan Zheng return ERR_PTR(-ENOMEM); 1142e02119d5SChris Mason 1143e02119d5SChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 1144e02119d5SChris Mason tree_root->sectorsize, tree_root->stripesize, 1145e02119d5SChris Mason root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1146e02119d5SChris Mason 1147e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1148e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1149e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 11507237f183SYan Zheng /* 11517237f183SYan Zheng * log trees do not get reference counted because they go away 11527237f183SYan Zheng * before a real commit is actually done. They do store pointers 11537237f183SYan Zheng * to file data extents, and those reference counts still get 11547237f183SYan Zheng * updated (along with back refs to the log tree). 11557237f183SYan Zheng */ 1156e02119d5SChris Mason root->ref_cows = 0; 1157e02119d5SChris Mason 11585d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 11595d4f98a2SYan Zheng BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0); 11607237f183SYan Zheng if (IS_ERR(leaf)) { 11617237f183SYan Zheng kfree(root); 11627237f183SYan Zheng return ERR_CAST(leaf); 11637237f183SYan Zheng } 1164e02119d5SChris Mason 11655d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 11665d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 11675d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 11685d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 11695d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 11707237f183SYan Zheng root->node = leaf; 1171e02119d5SChris Mason 1172e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 1173e02119d5SChris Mason (unsigned long)btrfs_header_fsid(root->node), 1174e02119d5SChris Mason BTRFS_FSID_SIZE); 1175e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1176e02119d5SChris Mason btrfs_tree_unlock(root->node); 11777237f183SYan Zheng return root; 11787237f183SYan Zheng } 11797237f183SYan Zheng 11807237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 11817237f183SYan Zheng struct btrfs_fs_info *fs_info) 11827237f183SYan Zheng { 11837237f183SYan Zheng struct btrfs_root *log_root; 11847237f183SYan Zheng 11857237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 11867237f183SYan Zheng if (IS_ERR(log_root)) 11877237f183SYan Zheng return PTR_ERR(log_root); 11887237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 11897237f183SYan Zheng fs_info->log_root_tree = log_root; 11907237f183SYan Zheng return 0; 11917237f183SYan Zheng } 11927237f183SYan Zheng 11937237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 11947237f183SYan Zheng struct btrfs_root *root) 11957237f183SYan Zheng { 11967237f183SYan Zheng struct btrfs_root *log_root; 11977237f183SYan Zheng struct btrfs_inode_item *inode_item; 11987237f183SYan Zheng 11997237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 12007237f183SYan Zheng if (IS_ERR(log_root)) 12017237f183SYan Zheng return PTR_ERR(log_root); 12027237f183SYan Zheng 12037237f183SYan Zheng log_root->last_trans = trans->transid; 12047237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 12057237f183SYan Zheng 12067237f183SYan Zheng inode_item = &log_root->root_item.inode; 12077237f183SYan Zheng inode_item->generation = cpu_to_le64(1); 12087237f183SYan Zheng inode_item->size = cpu_to_le64(3); 12097237f183SYan Zheng inode_item->nlink = cpu_to_le32(1); 12107237f183SYan Zheng inode_item->nbytes = cpu_to_le64(root->leafsize); 12117237f183SYan Zheng inode_item->mode = cpu_to_le32(S_IFDIR | 0755); 12127237f183SYan Zheng 12135d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 12147237f183SYan Zheng 12157237f183SYan Zheng WARN_ON(root->log_root); 12167237f183SYan Zheng root->log_root = log_root; 12177237f183SYan Zheng root->log_transid = 0; 1218257c62e1SChris Mason root->last_log_commit = 0; 1219e02119d5SChris Mason return 0; 1220e02119d5SChris Mason } 1221e02119d5SChris Mason 1222e02119d5SChris Mason struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root, 1223e02119d5SChris Mason struct btrfs_key *location) 1224e02119d5SChris Mason { 1225e02119d5SChris Mason struct btrfs_root *root; 1226e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 12270f7d52f4SChris Mason struct btrfs_path *path; 12285f39d397SChris Mason struct extent_buffer *l; 122984234f3aSYan Zheng u64 generation; 1230db94535dSChris Mason u32 blocksize; 12310f7d52f4SChris Mason int ret = 0; 12320f7d52f4SChris Mason 12335eda7b5eSChris Mason root = kzalloc(sizeof(*root), GFP_NOFS); 12340cf6c620SChris Mason if (!root) 12350f7d52f4SChris Mason return ERR_PTR(-ENOMEM); 12360f7d52f4SChris Mason if (location->offset == (u64)-1) { 1237db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 12380f7d52f4SChris Mason location->objectid, root); 12390f7d52f4SChris Mason if (ret) { 12400f7d52f4SChris Mason kfree(root); 12410f7d52f4SChris Mason return ERR_PTR(ret); 12420f7d52f4SChris Mason } 124313a8a7c8SYan, Zheng goto out; 12440f7d52f4SChris Mason } 12450f7d52f4SChris Mason 1246db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 124787ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 124887ee04ebSChris Mason root, fs_info, location->objectid); 12490f7d52f4SChris Mason 12500f7d52f4SChris Mason path = btrfs_alloc_path(); 12510f7d52f4SChris Mason BUG_ON(!path); 12520f7d52f4SChris Mason ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0); 125313a8a7c8SYan, Zheng if (ret == 0) { 12545f39d397SChris Mason l = path->nodes[0]; 12555f39d397SChris Mason read_extent_buffer(l, &root->root_item, 12565f39d397SChris Mason btrfs_item_ptr_offset(l, path->slots[0]), 12570f7d52f4SChris Mason sizeof(root->root_item)); 125844b36eb2SYan Zheng memcpy(&root->root_key, location, sizeof(*location)); 125913a8a7c8SYan, Zheng } 12600f7d52f4SChris Mason btrfs_free_path(path); 12610f7d52f4SChris Mason if (ret) { 12625e540f77STsutomu Itoh kfree(root); 126313a8a7c8SYan, Zheng if (ret > 0) 126413a8a7c8SYan, Zheng ret = -ENOENT; 12650f7d52f4SChris Mason return ERR_PTR(ret); 12660f7d52f4SChris Mason } 126713a8a7c8SYan, Zheng 126884234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1269db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1270db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 127184234f3aSYan Zheng blocksize, generation); 12725d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 12730f7d52f4SChris Mason BUG_ON(!root->node); 127413a8a7c8SYan, Zheng out: 127513a8a7c8SYan, Zheng if (location->objectid != BTRFS_TREE_LOG_OBJECTID) 12760f7d52f4SChris Mason root->ref_cows = 1; 127713a8a7c8SYan, Zheng 12785eda7b5eSChris Mason return root; 12795eda7b5eSChris Mason } 12805eda7b5eSChris Mason 1281dc17ff8fSChris Mason struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1282dc17ff8fSChris Mason u64 root_objectid) 1283dc17ff8fSChris Mason { 1284dc17ff8fSChris Mason struct btrfs_root *root; 1285dc17ff8fSChris Mason 1286dc17ff8fSChris Mason if (root_objectid == BTRFS_ROOT_TREE_OBJECTID) 1287dc17ff8fSChris Mason return fs_info->tree_root; 1288dc17ff8fSChris Mason if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) 1289dc17ff8fSChris Mason return fs_info->extent_root; 1290dc17ff8fSChris Mason 1291dc17ff8fSChris Mason root = radix_tree_lookup(&fs_info->fs_roots_radix, 1292dc17ff8fSChris Mason (unsigned long)root_objectid); 1293dc17ff8fSChris Mason return root; 1294dc17ff8fSChris Mason } 1295dc17ff8fSChris Mason 1296edbd8d4eSChris Mason struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info, 1297edbd8d4eSChris Mason struct btrfs_key *location) 12985eda7b5eSChris Mason { 12995eda7b5eSChris Mason struct btrfs_root *root; 13005eda7b5eSChris Mason int ret; 13015eda7b5eSChris Mason 1302edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1303edbd8d4eSChris Mason return fs_info->tree_root; 1304edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1305edbd8d4eSChris Mason return fs_info->extent_root; 13068f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 13078f18cf13SChris Mason return fs_info->chunk_root; 13088f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 13098f18cf13SChris Mason return fs_info->dev_root; 13100403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 13110403e47eSYan Zheng return fs_info->csum_root; 13124df27c4dSYan, Zheng again: 13134df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 13145eda7b5eSChris Mason root = radix_tree_lookup(&fs_info->fs_roots_radix, 13155eda7b5eSChris Mason (unsigned long)location->objectid); 13164df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 13175eda7b5eSChris Mason if (root) 13185eda7b5eSChris Mason return root; 13195eda7b5eSChris Mason 1320e02119d5SChris Mason root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location); 13215eda7b5eSChris Mason if (IS_ERR(root)) 13225eda7b5eSChris Mason return root; 13233394e160SChris Mason 13243394e160SChris Mason set_anon_super(&root->anon_super, NULL); 13253394e160SChris Mason 1326d68fc57bSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) { 1327d68fc57bSYan, Zheng ret = -ENOENT; 1328d68fc57bSYan, Zheng goto fail; 1329d68fc57bSYan, Zheng } 1330d68fc57bSYan, Zheng 1331d68fc57bSYan, Zheng ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid); 1332d68fc57bSYan, Zheng if (ret < 0) 1333d68fc57bSYan, Zheng goto fail; 1334d68fc57bSYan, Zheng if (ret == 0) 1335d68fc57bSYan, Zheng root->orphan_item_inserted = 1; 1336d68fc57bSYan, Zheng 13374df27c4dSYan, Zheng ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 13384df27c4dSYan, Zheng if (ret) 13394df27c4dSYan, Zheng goto fail; 13404df27c4dSYan, Zheng 13414df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 13422619ba1fSChris Mason ret = radix_tree_insert(&fs_info->fs_roots_radix, 13432619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13440f7d52f4SChris Mason root); 1345d68fc57bSYan, Zheng if (ret == 0) 13464df27c4dSYan, Zheng root->in_radix = 1; 1347d68fc57bSYan, Zheng 13484df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 13494df27c4dSYan, Zheng radix_tree_preload_end(); 13500f7d52f4SChris Mason if (ret) { 13514df27c4dSYan, Zheng if (ret == -EEXIST) { 13524df27c4dSYan, Zheng free_fs_root(root); 13534df27c4dSYan, Zheng goto again; 13540f7d52f4SChris Mason } 13554df27c4dSYan, Zheng goto fail; 13564df27c4dSYan, Zheng } 13574df27c4dSYan, Zheng 1358edbd8d4eSChris Mason ret = btrfs_find_dead_roots(fs_info->tree_root, 13595d4f98a2SYan Zheng root->root_key.objectid); 13604df27c4dSYan, Zheng WARN_ON(ret); 1361edbd8d4eSChris Mason return root; 13624df27c4dSYan, Zheng fail: 13634df27c4dSYan, Zheng free_fs_root(root); 13644df27c4dSYan, Zheng return ERR_PTR(ret); 1365edbd8d4eSChris Mason } 1366edbd8d4eSChris Mason 1367edbd8d4eSChris Mason struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info, 1368edbd8d4eSChris Mason struct btrfs_key *location, 1369edbd8d4eSChris Mason const char *name, int namelen) 1370edbd8d4eSChris Mason { 13714df27c4dSYan, Zheng return btrfs_read_fs_root_no_name(fs_info, location); 13724df27c4dSYan, Zheng #if 0 1373edbd8d4eSChris Mason struct btrfs_root *root; 1374edbd8d4eSChris Mason int ret; 1375edbd8d4eSChris Mason 1376edbd8d4eSChris Mason root = btrfs_read_fs_root_no_name(fs_info, location); 1377edbd8d4eSChris Mason if (!root) 1378edbd8d4eSChris Mason return NULL; 137958176a96SJosef Bacik 13804313b399SChris Mason if (root->in_sysfs) 13814313b399SChris Mason return root; 13824313b399SChris Mason 138358176a96SJosef Bacik ret = btrfs_set_root_name(root, name, namelen); 138458176a96SJosef Bacik if (ret) { 13855f39d397SChris Mason free_extent_buffer(root->node); 138658176a96SJosef Bacik kfree(root); 138758176a96SJosef Bacik return ERR_PTR(ret); 138858176a96SJosef Bacik } 13894df27c4dSYan, Zheng 139058176a96SJosef Bacik ret = btrfs_sysfs_add_root(root); 139158176a96SJosef Bacik if (ret) { 13925f39d397SChris Mason free_extent_buffer(root->node); 139358176a96SJosef Bacik kfree(root->name); 139458176a96SJosef Bacik kfree(root); 139558176a96SJosef Bacik return ERR_PTR(ret); 139658176a96SJosef Bacik } 13974313b399SChris Mason root->in_sysfs = 1; 13980f7d52f4SChris Mason return root; 13994df27c4dSYan, Zheng #endif 14000f7d52f4SChris Mason } 140104160088SChris Mason 140204160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 140304160088SChris Mason { 140404160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 140504160088SChris Mason int ret = 0; 140604160088SChris Mason struct btrfs_device *device; 140704160088SChris Mason struct backing_dev_info *bdi; 1408b7967db7SChris Mason 1409c6e30871SQinghuang Feng list_for_each_entry(device, &info->fs_devices->devices, dev_list) { 1410dfe25020SChris Mason if (!device->bdev) 1411dfe25020SChris Mason continue; 141204160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 141304160088SChris Mason if (bdi && bdi_congested(bdi, bdi_bits)) { 141404160088SChris Mason ret = 1; 141504160088SChris Mason break; 141604160088SChris Mason } 141704160088SChris Mason } 141804160088SChris Mason return ret; 141904160088SChris Mason } 142004160088SChris Mason 142138b66988SChris Mason /* 142238b66988SChris Mason * this unplugs every device on the box, and it is only used when page 142338b66988SChris Mason * is null 142438b66988SChris Mason */ 142538b66988SChris Mason static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 142638b66988SChris Mason { 142738b66988SChris Mason struct btrfs_device *device; 142838b66988SChris Mason struct btrfs_fs_info *info; 142938b66988SChris Mason 143038b66988SChris Mason info = (struct btrfs_fs_info *)bdi->unplug_io_data; 1431c6e30871SQinghuang Feng list_for_each_entry(device, &info->fs_devices->devices, dev_list) { 1432d20f7043SChris Mason if (!device->bdev) 1433d20f7043SChris Mason continue; 1434d20f7043SChris Mason 143538b66988SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 1436d397712bSChris Mason if (bdi->unplug_io_fn) 143738b66988SChris Mason bdi->unplug_io_fn(bdi, page); 143838b66988SChris Mason } 143938b66988SChris Mason } 144038b66988SChris Mason 1441b2950863SChristoph Hellwig static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) 144204160088SChris Mason { 144338b66988SChris Mason struct inode *inode; 1444f2d8d74dSChris Mason struct extent_map_tree *em_tree; 1445f2d8d74dSChris Mason struct extent_map *em; 1446bcbfce8aSChris Mason struct address_space *mapping; 144738b66988SChris Mason u64 offset; 144838b66988SChris Mason 1449bcbfce8aSChris Mason /* the generic O_DIRECT read code does this */ 14509f0ba5bdSChris Mason if (1 || !page) { 145138b66988SChris Mason __unplug_io_fn(bdi, page); 145238b66988SChris Mason return; 145338b66988SChris Mason } 145438b66988SChris Mason 1455bcbfce8aSChris Mason /* 1456bcbfce8aSChris Mason * page->mapping may change at any time. Get a consistent copy 1457bcbfce8aSChris Mason * and use that for everything below 1458bcbfce8aSChris Mason */ 1459bcbfce8aSChris Mason smp_mb(); 1460bcbfce8aSChris Mason mapping = page->mapping; 1461bcbfce8aSChris Mason if (!mapping) 1462bcbfce8aSChris Mason return; 1463bcbfce8aSChris Mason 1464bcbfce8aSChris Mason inode = mapping->host; 1465240d5d48SChris Mason 1466240d5d48SChris Mason /* 1467240d5d48SChris Mason * don't do the expensive searching for a small number of 1468240d5d48SChris Mason * devices 1469240d5d48SChris Mason */ 1470240d5d48SChris Mason if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) { 1471240d5d48SChris Mason __unplug_io_fn(bdi, page); 1472240d5d48SChris Mason return; 1473240d5d48SChris Mason } 1474240d5d48SChris Mason 147538b66988SChris Mason offset = page_offset(page); 147604160088SChris Mason 1477f2d8d74dSChris Mason em_tree = &BTRFS_I(inode)->extent_tree; 1478890871beSChris Mason read_lock(&em_tree->lock); 1479f2d8d74dSChris Mason em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE); 1480890871beSChris Mason read_unlock(&em_tree->lock); 148189642229SChris Mason if (!em) { 148289642229SChris Mason __unplug_io_fn(bdi, page); 1483f2d8d74dSChris Mason return; 148489642229SChris Mason } 1485f2d8d74dSChris Mason 148689642229SChris Mason if (em->block_start >= EXTENT_MAP_LAST_BYTE) { 148789642229SChris Mason free_extent_map(em); 148889642229SChris Mason __unplug_io_fn(bdi, page); 148989642229SChris Mason return; 149089642229SChris Mason } 1491f2d8d74dSChris Mason offset = offset - em->start; 1492f2d8d74dSChris Mason btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree, 1493f2d8d74dSChris Mason em->block_start + offset, page); 1494f2d8d74dSChris Mason free_extent_map(em); 149504160088SChris Mason } 149604160088SChris Mason 1497ad081f14SJens Axboe /* 1498ad081f14SJens Axboe * If this fails, caller must call bdi_destroy() to get rid of the 1499ad081f14SJens Axboe * bdi again. 1500ad081f14SJens Axboe */ 150104160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 150204160088SChris Mason { 1503ad081f14SJens Axboe int err; 1504ad081f14SJens Axboe 1505ad081f14SJens Axboe bdi->capabilities = BDI_CAP_MAP_COPY; 1506e6d086d8SJens Axboe err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1507ad081f14SJens Axboe if (err) 1508ad081f14SJens Axboe return err; 1509ad081f14SJens Axboe 15104575c9ccSChris Mason bdi->ra_pages = default_backing_dev_info.ra_pages; 151104160088SChris Mason bdi->unplug_io_fn = btrfs_unplug_io_fn; 151204160088SChris Mason bdi->unplug_io_data = info; 151304160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 151404160088SChris Mason bdi->congested_data = info; 151504160088SChris Mason return 0; 151604160088SChris Mason } 151704160088SChris Mason 1518ce9adaa5SChris Mason static int bio_ready_for_csum(struct bio *bio) 1519ce9adaa5SChris Mason { 1520ce9adaa5SChris Mason u64 length = 0; 1521ce9adaa5SChris Mason u64 buf_len = 0; 1522ce9adaa5SChris Mason u64 start = 0; 1523ce9adaa5SChris Mason struct page *page; 1524ce9adaa5SChris Mason struct extent_io_tree *io_tree = NULL; 1525ce9adaa5SChris Mason struct bio_vec *bvec; 1526ce9adaa5SChris Mason int i; 1527ce9adaa5SChris Mason int ret; 1528ce9adaa5SChris Mason 1529ce9adaa5SChris Mason bio_for_each_segment(bvec, bio, i) { 1530ce9adaa5SChris Mason page = bvec->bv_page; 1531ce9adaa5SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) { 1532ce9adaa5SChris Mason length += bvec->bv_len; 1533ce9adaa5SChris Mason continue; 1534ce9adaa5SChris Mason } 1535ce9adaa5SChris Mason if (!page->private) { 1536ce9adaa5SChris Mason length += bvec->bv_len; 1537ce9adaa5SChris Mason continue; 1538ce9adaa5SChris Mason } 1539ce9adaa5SChris Mason length = bvec->bv_len; 1540ce9adaa5SChris Mason buf_len = page->private >> 2; 1541ce9adaa5SChris Mason start = page_offset(page) + bvec->bv_offset; 1542ce9adaa5SChris Mason io_tree = &BTRFS_I(page->mapping->host)->io_tree; 1543ce9adaa5SChris Mason } 1544ce9adaa5SChris Mason /* are we fully contained in this bio? */ 1545ce9adaa5SChris Mason if (buf_len <= length) 1546ce9adaa5SChris Mason return 1; 1547ce9adaa5SChris Mason 1548ce9adaa5SChris Mason ret = extent_range_uptodate(io_tree, start + length, 1549ce9adaa5SChris Mason start + buf_len - 1); 1550ce9adaa5SChris Mason return ret; 1551ce9adaa5SChris Mason } 1552ce9adaa5SChris Mason 15538b712842SChris Mason /* 15548b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 15558b712842SChris Mason * functions. This is where read checksum verification actually happens 15568b712842SChris Mason */ 15578b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1558ce9adaa5SChris Mason { 1559ce9adaa5SChris Mason struct bio *bio; 15608b712842SChris Mason struct end_io_wq *end_io_wq; 15618b712842SChris Mason struct btrfs_fs_info *fs_info; 1562ce9adaa5SChris Mason int error; 1563ce9adaa5SChris Mason 15648b712842SChris Mason end_io_wq = container_of(work, struct end_io_wq, work); 1565ce9adaa5SChris Mason bio = end_io_wq->bio; 15668b712842SChris Mason fs_info = end_io_wq->info; 15678b712842SChris Mason 1568cad321adSChris Mason /* metadata bio reads are special because the whole tree block must 15698b712842SChris Mason * be checksummed at once. This makes sure the entire block is in 15708b712842SChris Mason * ram and up to date before trying to verify things. For 15718b712842SChris Mason * blocksize <= pagesize, it is basically a noop 15728b712842SChris Mason */ 15737b6d91daSChristoph Hellwig if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata && 1574cad321adSChris Mason !bio_ready_for_csum(bio)) { 1575d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 15768b712842SChris Mason &end_io_wq->work); 1577ce9adaa5SChris Mason return; 1578ce9adaa5SChris Mason } 1579ce9adaa5SChris Mason error = end_io_wq->error; 1580ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1581ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 1582ce9adaa5SChris Mason kfree(end_io_wq); 1583ce9adaa5SChris Mason bio_endio(bio, error); 1584ce9adaa5SChris Mason } 158544b8bd7eSChris Mason 1586a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1587a74a4b97SChris Mason { 1588a74a4b97SChris Mason struct btrfs_root *root = arg; 1589a74a4b97SChris Mason 1590a74a4b97SChris Mason do { 1591a74a4b97SChris Mason vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 159276dda93cSYan, Zheng 159376dda93cSYan, Zheng if (!(root->fs_info->sb->s_flags & MS_RDONLY) && 159476dda93cSYan, Zheng mutex_trylock(&root->fs_info->cleaner_mutex)) { 159524bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 1596a74a4b97SChris Mason btrfs_clean_old_snapshots(root); 1597a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 159876dda93cSYan, Zheng } 1599a74a4b97SChris Mason 1600a74a4b97SChris Mason if (freezing(current)) { 1601a74a4b97SChris Mason refrigerator(); 1602a74a4b97SChris Mason } else { 1603a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 16048929ecfaSYan, Zheng if (!kthread_should_stop()) 1605a74a4b97SChris Mason schedule(); 1606a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1607a74a4b97SChris Mason } 1608a74a4b97SChris Mason } while (!kthread_should_stop()); 1609a74a4b97SChris Mason return 0; 1610a74a4b97SChris Mason } 1611a74a4b97SChris Mason 1612a74a4b97SChris Mason static int transaction_kthread(void *arg) 1613a74a4b97SChris Mason { 1614a74a4b97SChris Mason struct btrfs_root *root = arg; 1615a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1616a74a4b97SChris Mason struct btrfs_transaction *cur; 16178929ecfaSYan, Zheng u64 transid; 1618a74a4b97SChris Mason unsigned long now; 1619a74a4b97SChris Mason unsigned long delay; 1620a74a4b97SChris Mason int ret; 1621a74a4b97SChris Mason 1622a74a4b97SChris Mason do { 1623a74a4b97SChris Mason delay = HZ * 30; 1624a74a4b97SChris Mason vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1625a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1626a74a4b97SChris Mason 16278929ecfaSYan, Zheng spin_lock(&root->fs_info->new_trans_lock); 1628a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1629a74a4b97SChris Mason if (!cur) { 16308929ecfaSYan, Zheng spin_unlock(&root->fs_info->new_trans_lock); 1631a74a4b97SChris Mason goto sleep; 1632a74a4b97SChris Mason } 163331153d81SYan Zheng 1634a74a4b97SChris Mason now = get_seconds(); 16358929ecfaSYan, Zheng if (!cur->blocked && 16368929ecfaSYan, Zheng (now < cur->start_time || now - cur->start_time < 30)) { 16378929ecfaSYan, Zheng spin_unlock(&root->fs_info->new_trans_lock); 1638a74a4b97SChris Mason delay = HZ * 5; 1639a74a4b97SChris Mason goto sleep; 1640a74a4b97SChris Mason } 16418929ecfaSYan, Zheng transid = cur->transid; 16428929ecfaSYan, Zheng spin_unlock(&root->fs_info->new_trans_lock); 164356bec294SChris Mason 16448929ecfaSYan, Zheng trans = btrfs_join_transaction(root, 1); 16453612b495STsutomu Itoh BUG_ON(IS_ERR(trans)); 16468929ecfaSYan, Zheng if (transid == trans->transid) { 16478929ecfaSYan, Zheng ret = btrfs_commit_transaction(trans, root); 16488929ecfaSYan, Zheng BUG_ON(ret); 16498929ecfaSYan, Zheng } else { 16508929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 16518929ecfaSYan, Zheng } 1652a74a4b97SChris Mason sleep: 1653a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1654a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1655a74a4b97SChris Mason 1656a74a4b97SChris Mason if (freezing(current)) { 1657a74a4b97SChris Mason refrigerator(); 1658a74a4b97SChris Mason } else { 1659a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 16608929ecfaSYan, Zheng if (!kthread_should_stop() && 16618929ecfaSYan, Zheng !btrfs_transaction_blocked(root->fs_info)) 1662a74a4b97SChris Mason schedule_timeout(delay); 1663a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1664a74a4b97SChris Mason } 1665a74a4b97SChris Mason } while (!kthread_should_stop()); 1666a74a4b97SChris Mason return 0; 1667a74a4b97SChris Mason } 1668a74a4b97SChris Mason 16698a4b83ccSChris Mason struct btrfs_root *open_ctree(struct super_block *sb, 1670dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 1671dfe25020SChris Mason char *options) 1672eb60ceacSChris Mason { 1673db94535dSChris Mason u32 sectorsize; 1674db94535dSChris Mason u32 nodesize; 1675db94535dSChris Mason u32 leafsize; 1676db94535dSChris Mason u32 blocksize; 167787ee04ebSChris Mason u32 stripesize; 167884234f3aSYan Zheng u64 generation; 1679f2b636e8SJosef Bacik u64 features; 16803de4586cSChris Mason struct btrfs_key location; 1681a061fc8dSChris Mason struct buffer_head *bh; 1682e02119d5SChris Mason struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root), 1683e20d96d6SChris Mason GFP_NOFS); 1684d20f7043SChris Mason struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root), 1685d20f7043SChris Mason GFP_NOFS); 1686450ba0eaSJosef Bacik struct btrfs_root *tree_root = btrfs_sb(sb); 1687450ba0eaSJosef Bacik struct btrfs_fs_info *fs_info = tree_root->fs_info; 1688e02119d5SChris Mason struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root), 16890b86a832SChris Mason GFP_NOFS); 1690e02119d5SChris Mason struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root), 16910b86a832SChris Mason GFP_NOFS); 1692e02119d5SChris Mason struct btrfs_root *log_tree_root; 1693e02119d5SChris Mason 1694eb60ceacSChris Mason int ret; 1695e58ca020SYan int err = -EINVAL; 16964543df7eSChris Mason 16972c90e5d6SChris Mason struct btrfs_super_block *disk_super; 16988790d502SChris Mason 16990463bb4eSJim Meyering if (!extent_root || !tree_root || !fs_info || 1700d20f7043SChris Mason !chunk_root || !dev_root || !csum_root) { 170139279cc3SChris Mason err = -ENOMEM; 170239279cc3SChris Mason goto fail; 170339279cc3SChris Mason } 170476dda93cSYan, Zheng 170576dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 170676dda93cSYan, Zheng if (ret) { 170776dda93cSYan, Zheng err = ret; 170876dda93cSYan, Zheng goto fail; 170976dda93cSYan, Zheng } 171076dda93cSYan, Zheng 171176dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 171276dda93cSYan, Zheng if (ret) { 171376dda93cSYan, Zheng err = ret; 171476dda93cSYan, Zheng goto fail_srcu; 171576dda93cSYan, Zheng } 171676dda93cSYan, Zheng 171776dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 171876dda93cSYan, Zheng if (!fs_info->btree_inode) { 171976dda93cSYan, Zheng err = -ENOMEM; 172076dda93cSYan, Zheng goto fail_bdi; 172176dda93cSYan, Zheng } 172276dda93cSYan, Zheng 172376dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 17248fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 1725facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 172624bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 172719c00ddcSChris Mason INIT_LIST_HEAD(&fs_info->hashers); 1728ea8c2819SChris Mason INIT_LIST_HEAD(&fs_info->delalloc_inodes); 17295a3f23d5SChris Mason INIT_LIST_HEAD(&fs_info->ordered_operations); 173011833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 17311832a6d5SChris Mason spin_lock_init(&fs_info->delalloc_lock); 1732cee36a03SChris Mason spin_lock_init(&fs_info->new_trans_lock); 173331153d81SYan Zheng spin_lock_init(&fs_info->ref_cache_lock); 173476dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 173524bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 173619c00ddcSChris Mason 173758176a96SJosef Bacik init_completion(&fs_info->kobj_unregister); 17389f5fae2fSChris Mason fs_info->tree_root = tree_root; 17399f5fae2fSChris Mason fs_info->extent_root = extent_root; 1740d20f7043SChris Mason fs_info->csum_root = csum_root; 17410b86a832SChris Mason fs_info->chunk_root = chunk_root; 17420b86a832SChris Mason fs_info->dev_root = dev_root; 17438a4b83ccSChris Mason fs_info->fs_devices = fs_devices; 17440b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 17456324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 17460b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 1747f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->global_block_rsv); 1748f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->delalloc_block_rsv); 1749f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->trans_block_rsv); 1750f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->chunk_block_rsv); 1751f0486c68SYan, Zheng btrfs_init_block_rsv(&fs_info->empty_block_rsv); 1752f0486c68SYan, Zheng INIT_LIST_HEAD(&fs_info->durable_block_rsv_list); 1753f0486c68SYan, Zheng mutex_init(&fs_info->durable_block_rsv_mutex); 1754cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 1755771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 17568c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 17570986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 1758e20d96d6SChris Mason fs_info->sb = sb; 17596f568d35SChris Mason fs_info->max_inline = 8192 * 1024; 17609ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 1761c8b97818SChris Mason 1762b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 1763b34b086cSChris Mason num_online_cpus() + 2, 8); 17640afbaf8cSChris Mason 17653eaa2885SChris Mason INIT_LIST_HEAD(&fs_info->ordered_extents); 17663eaa2885SChris Mason spin_lock_init(&fs_info->ordered_extent_lock); 17673eaa2885SChris Mason 1768a061fc8dSChris Mason sb->s_blocksize = 4096; 1769a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 177032a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 1771a061fc8dSChris Mason 177276dda93cSYan, Zheng fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 177376dda93cSYan, Zheng fs_info->btree_inode->i_nlink = 1; 17740afbaf8cSChris Mason /* 17750afbaf8cSChris Mason * we set the i_size on the btree inode to the max possible int. 17760afbaf8cSChris Mason * the real end of the address space is determined by all of 17770afbaf8cSChris Mason * the devices in the system 17780afbaf8cSChris Mason */ 17790afbaf8cSChris Mason fs_info->btree_inode->i_size = OFFSET_MAX; 1780d98237b3SChris Mason fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 178104160088SChris Mason fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 178204160088SChris Mason 17835d4f98a2SYan Zheng RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 1784d1310b2eSChris Mason extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 17855f39d397SChris Mason fs_info->btree_inode->i_mapping, 17865f39d397SChris Mason GFP_NOFS); 1787d1310b2eSChris Mason extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree, 1788d1310b2eSChris Mason GFP_NOFS); 17890da5468fSChris Mason 1790d1310b2eSChris Mason BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 1791d1310b2eSChris Mason 17920f7d52f4SChris Mason BTRFS_I(fs_info->btree_inode)->root = tree_root; 17930f7d52f4SChris Mason memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 17940f7d52f4SChris Mason sizeof(struct btrfs_key)); 179576dda93cSYan, Zheng BTRFS_I(fs_info->btree_inode)->dummy_inode = 1; 179622b0ebdaSChris Mason insert_inode_hash(fs_info->btree_inode); 179739279cc3SChris Mason 1798e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 17996bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 1800925baeddSChris Mason 180111833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 18027d9eb12cSChris Mason fs_info->btree_inode->i_mapping, GFP_NOFS); 180311833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 180411833d66SYan Zheng fs_info->btree_inode->i_mapping, GFP_NOFS); 180511833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 1806e6dcd2dcSChris Mason fs_info->do_barriers = 1; 1807f9295749SChris Mason 1808d98237b3SChris Mason 1809509659cdSChris Mason mutex_init(&fs_info->trans_mutex); 18105a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 181130ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 1812925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 1813925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 1814a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 1815a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 1816276e680dSYan Zheng init_rwsem(&fs_info->extent_commit_sem); 1817c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 181876dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 1819fa9c0d79SChris Mason 1820fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 1821fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 1822fa9c0d79SChris Mason 18233768f368SChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 182419c00ddcSChris Mason init_waitqueue_head(&fs_info->transaction_wait); 1825bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 18264854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 18279a8dd150SChris Mason 18280b86a832SChris Mason __setup_root(4096, 4096, 4096, 4096, tree_root, 18292c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 18307eccb903SChris Mason 1831a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 183220b45077SDave Young if (!bh) { 183320b45077SDave Young err = -EINVAL; 183439279cc3SChris Mason goto fail_iput; 183520b45077SDave Young } 183639279cc3SChris Mason 1837a061fc8dSChris Mason memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy)); 18382d69a0f8SYan Zheng memcpy(&fs_info->super_for_commit, &fs_info->super_copy, 18392d69a0f8SYan Zheng sizeof(fs_info->super_for_commit)); 1840a061fc8dSChris Mason brelse(bh); 18415f39d397SChris Mason 1842a061fc8dSChris Mason memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE); 18430b86a832SChris Mason 18445f39d397SChris Mason disk_super = &fs_info->super_copy; 18450f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 1846c6e2bac1SJosef Bacik goto fail_iput; 18470f7d52f4SChris Mason 1848acce952bSliubo /* check FS state, whether FS is broken. */ 1849acce952bSliubo fs_info->fs_state |= btrfs_super_flags(disk_super); 1850acce952bSliubo 1851acce952bSliubo btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 1852acce952bSliubo 18532b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 18542b82032cSYan Zheng if (ret) { 18552b82032cSYan Zheng err = ret; 1856c6e2bac1SJosef Bacik goto fail_iput; 18572b82032cSYan Zheng } 1858dfe25020SChris Mason 1859f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 1860f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 1861f2b636e8SJosef Bacik if (features) { 1862f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 1863f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 186421380931SJoel Becker (unsigned long long)features); 1865f2b636e8SJosef Bacik err = -EINVAL; 1866c6e2bac1SJosef Bacik goto fail_iput; 1867f2b636e8SJosef Bacik } 1868f2b636e8SJosef Bacik 18695d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 18705d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 1871a6fa6faeSLi Zefan if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO) 1872a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 18735d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 18745d4f98a2SYan Zheng 1875f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 1876f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 1877f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 1878f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 1879f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 188021380931SJoel Becker (unsigned long long)features); 1881f2b636e8SJosef Bacik err = -EINVAL; 1882c6e2bac1SJosef Bacik goto fail_iput; 1883f2b636e8SJosef Bacik } 188461d92c32SChris Mason 188561d92c32SChris Mason btrfs_init_workers(&fs_info->generic_worker, 188661d92c32SChris Mason "genwork", 1, NULL); 188761d92c32SChris Mason 18885443be45SChris Mason btrfs_init_workers(&fs_info->workers, "worker", 188961d92c32SChris Mason fs_info->thread_pool_size, 189061d92c32SChris Mason &fs_info->generic_worker); 1891c8b97818SChris Mason 1892771ed689SChris Mason btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 189361d92c32SChris Mason fs_info->thread_pool_size, 189461d92c32SChris Mason &fs_info->generic_worker); 1895771ed689SChris Mason 18965443be45SChris Mason btrfs_init_workers(&fs_info->submit_workers, "submit", 1897b720d209SChris Mason min_t(u64, fs_devices->num_devices, 189861d92c32SChris Mason fs_info->thread_pool_size), 189961d92c32SChris Mason &fs_info->generic_worker); 190061b49440SChris Mason 190161b49440SChris Mason /* a higher idle thresh on the submit workers makes it much more 190261b49440SChris Mason * likely that bios will be send down in a sane order to the 190361b49440SChris Mason * devices 190461b49440SChris Mason */ 190561b49440SChris Mason fs_info->submit_workers.idle_thresh = 64; 190653863232SChris Mason 1907771ed689SChris Mason fs_info->workers.idle_thresh = 16; 19084a69a410SChris Mason fs_info->workers.ordered = 1; 190961b49440SChris Mason 1910771ed689SChris Mason fs_info->delalloc_workers.idle_thresh = 2; 1911771ed689SChris Mason fs_info->delalloc_workers.ordered = 1; 1912771ed689SChris Mason 191361d92c32SChris Mason btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1, 191461d92c32SChris Mason &fs_info->generic_worker); 19155443be45SChris Mason btrfs_init_workers(&fs_info->endio_workers, "endio", 191661d92c32SChris Mason fs_info->thread_pool_size, 191761d92c32SChris Mason &fs_info->generic_worker); 1918d20f7043SChris Mason btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 191961d92c32SChris Mason fs_info->thread_pool_size, 192061d92c32SChris Mason &fs_info->generic_worker); 1921cad321adSChris Mason btrfs_init_workers(&fs_info->endio_meta_write_workers, 192261d92c32SChris Mason "endio-meta-write", fs_info->thread_pool_size, 192361d92c32SChris Mason &fs_info->generic_worker); 19245443be45SChris Mason btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 192561d92c32SChris Mason fs_info->thread_pool_size, 192661d92c32SChris Mason &fs_info->generic_worker); 19270cb59c99SJosef Bacik btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write", 19280cb59c99SJosef Bacik 1, &fs_info->generic_worker); 192961b49440SChris Mason 193061b49440SChris Mason /* 193161b49440SChris Mason * endios are largely parallel and should have a very 193261b49440SChris Mason * low idle thresh 193361b49440SChris Mason */ 193461b49440SChris Mason fs_info->endio_workers.idle_thresh = 4; 1935b51912c9SChris Mason fs_info->endio_meta_workers.idle_thresh = 4; 1936b51912c9SChris Mason 19379042846bSChris Mason fs_info->endio_write_workers.idle_thresh = 2; 19389042846bSChris Mason fs_info->endio_meta_write_workers.idle_thresh = 2; 19399042846bSChris Mason 19404543df7eSChris Mason btrfs_start_workers(&fs_info->workers, 1); 194161d92c32SChris Mason btrfs_start_workers(&fs_info->generic_worker, 1); 19421cc127b5SChris Mason btrfs_start_workers(&fs_info->submit_workers, 1); 1943771ed689SChris Mason btrfs_start_workers(&fs_info->delalloc_workers, 1); 1944247e743cSChris Mason btrfs_start_workers(&fs_info->fixup_workers, 1); 19459042846bSChris Mason btrfs_start_workers(&fs_info->endio_workers, 1); 19469042846bSChris Mason btrfs_start_workers(&fs_info->endio_meta_workers, 1); 19479042846bSChris Mason btrfs_start_workers(&fs_info->endio_meta_write_workers, 1); 19489042846bSChris Mason btrfs_start_workers(&fs_info->endio_write_workers, 1); 19490cb59c99SJosef Bacik btrfs_start_workers(&fs_info->endio_freespace_worker, 1); 19504543df7eSChris Mason 19514575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 1952c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 1953c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 19544575c9ccSChris Mason 1955db94535dSChris Mason nodesize = btrfs_super_nodesize(disk_super); 1956db94535dSChris Mason leafsize = btrfs_super_leafsize(disk_super); 1957db94535dSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 195887ee04ebSChris Mason stripesize = btrfs_super_stripesize(disk_super); 1959db94535dSChris Mason tree_root->nodesize = nodesize; 1960db94535dSChris Mason tree_root->leafsize = leafsize; 1961db94535dSChris Mason tree_root->sectorsize = sectorsize; 196287ee04ebSChris Mason tree_root->stripesize = stripesize; 1963a061fc8dSChris Mason 1964a061fc8dSChris Mason sb->s_blocksize = sectorsize; 1965a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 1966db94535dSChris Mason 196739279cc3SChris Mason if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, 196839279cc3SChris Mason sizeof(disk_super->magic))) { 1969d397712bSChris Mason printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id); 197039279cc3SChris Mason goto fail_sb_buffer; 197139279cc3SChris Mason } 197219c00ddcSChris Mason 1973925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 1974e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 1975925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 197684eed90fSChris Mason if (ret) { 1977d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read the system " 1978d397712bSChris Mason "array on %s\n", sb->s_id); 19795d4f98a2SYan Zheng goto fail_sb_buffer; 198084eed90fSChris Mason } 19810b86a832SChris Mason 19820b86a832SChris Mason blocksize = btrfs_level_size(tree_root, 19830b86a832SChris Mason btrfs_super_chunk_root_level(disk_super)); 198484234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 19850b86a832SChris Mason 19860b86a832SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 19870b86a832SChris Mason chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 19880b86a832SChris Mason 19890b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 19900b86a832SChris Mason btrfs_super_chunk_root(disk_super), 199184234f3aSYan Zheng blocksize, generation); 19920b86a832SChris Mason BUG_ON(!chunk_root->node); 199383121942SDavid Woodhouse if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 199483121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n", 199583121942SDavid Woodhouse sb->s_id); 199683121942SDavid Woodhouse goto fail_chunk_root; 199783121942SDavid Woodhouse } 19985d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 19995d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 20000b86a832SChris Mason 2001e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2002e17cade2SChris Mason (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node), 2003e17cade2SChris Mason BTRFS_UUID_SIZE); 2004e17cade2SChris Mason 2005925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 20060b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 2007925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 20082b82032cSYan Zheng if (ret) { 2009d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n", 2010d397712bSChris Mason sb->s_id); 20112b82032cSYan Zheng goto fail_chunk_root; 20122b82032cSYan Zheng } 20130b86a832SChris Mason 2014dfe25020SChris Mason btrfs_close_extra_devices(fs_devices); 2015dfe25020SChris Mason 2016db94535dSChris Mason blocksize = btrfs_level_size(tree_root, 2017db94535dSChris Mason btrfs_super_root_level(disk_super)); 201884234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 20190b86a832SChris Mason 2020e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 2021db94535dSChris Mason btrfs_super_root(disk_super), 202284234f3aSYan Zheng blocksize, generation); 202339279cc3SChris Mason if (!tree_root->node) 20242b82032cSYan Zheng goto fail_chunk_root; 202583121942SDavid Woodhouse if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 202683121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read tree root on %s\n", 202783121942SDavid Woodhouse sb->s_id); 202883121942SDavid Woodhouse goto fail_tree_root; 202983121942SDavid Woodhouse } 20305d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 20315d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 2032db94535dSChris Mason 2033db94535dSChris Mason ret = find_and_setup_root(tree_root, fs_info, 2034e20d96d6SChris Mason BTRFS_EXTENT_TREE_OBJECTID, extent_root); 20350b86a832SChris Mason if (ret) 203639279cc3SChris Mason goto fail_tree_root; 20370b86a832SChris Mason extent_root->track_dirty = 1; 20380b86a832SChris Mason 20390b86a832SChris Mason ret = find_and_setup_root(tree_root, fs_info, 20400b86a832SChris Mason BTRFS_DEV_TREE_OBJECTID, dev_root); 20410b86a832SChris Mason if (ret) 20420b86a832SChris Mason goto fail_extent_root; 20435d4f98a2SYan Zheng dev_root->track_dirty = 1; 20443768f368SChris Mason 2045d20f7043SChris Mason ret = find_and_setup_root(tree_root, fs_info, 2046d20f7043SChris Mason BTRFS_CSUM_TREE_OBJECTID, csum_root); 2047d20f7043SChris Mason if (ret) 20485d4f98a2SYan Zheng goto fail_dev_root; 2049d20f7043SChris Mason 2050d20f7043SChris Mason csum_root->track_dirty = 1; 2051d20f7043SChris Mason 20528929ecfaSYan, Zheng fs_info->generation = generation; 20538929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 20548929ecfaSYan, Zheng fs_info->data_alloc_profile = (u64)-1; 20558929ecfaSYan, Zheng fs_info->metadata_alloc_profile = (u64)-1; 20568929ecfaSYan, Zheng fs_info->system_alloc_profile = fs_info->metadata_alloc_profile; 20578929ecfaSYan, Zheng 20581b1d1f66SJosef Bacik ret = btrfs_read_block_groups(extent_root); 20591b1d1f66SJosef Bacik if (ret) { 20601b1d1f66SJosef Bacik printk(KERN_ERR "Failed to read block groups: %d\n", ret); 20611b1d1f66SJosef Bacik goto fail_block_groups; 20621b1d1f66SJosef Bacik } 20639078a3e1SChris Mason 2064a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2065a74a4b97SChris Mason "btrfs-cleaner"); 206657506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 20671b1d1f66SJosef Bacik goto fail_block_groups; 2068a74a4b97SChris Mason 2069a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 2070a74a4b97SChris Mason tree_root, 2071a74a4b97SChris Mason "btrfs-transaction"); 207257506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 20733f157a2fSChris Mason goto fail_cleaner; 2074a74a4b97SChris Mason 2075c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 2076c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 2077c289811cSChris Mason !fs_info->fs_devices->rotating) { 2078c289811cSChris Mason printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD " 2079c289811cSChris Mason "mode\n"); 2080c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2081c289811cSChris Mason } 2082c289811cSChris Mason 2083acce952bSliubo /* do not make disk changes in broken FS */ 2084acce952bSliubo if (btrfs_super_log_root(disk_super) != 0 && 2085acce952bSliubo !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) { 2086e02119d5SChris Mason u64 bytenr = btrfs_super_log_root(disk_super); 2087d18a2c44SChris Mason 20887c2ca468SChris Mason if (fs_devices->rw_devices == 0) { 2089d397712bSChris Mason printk(KERN_WARNING "Btrfs log replay required " 2090d397712bSChris Mason "on RO media\n"); 20917c2ca468SChris Mason err = -EIO; 20927c2ca468SChris Mason goto fail_trans_kthread; 20937c2ca468SChris Mason } 2094e02119d5SChris Mason blocksize = 2095e02119d5SChris Mason btrfs_level_size(tree_root, 2096e02119d5SChris Mason btrfs_super_log_root_level(disk_super)); 2097e02119d5SChris Mason 2098676e4c86SDan Carpenter log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS); 2099676e4c86SDan Carpenter if (!log_tree_root) { 2100676e4c86SDan Carpenter err = -ENOMEM; 2101676e4c86SDan Carpenter goto fail_trans_kthread; 2102676e4c86SDan Carpenter } 2103e02119d5SChris Mason 2104e02119d5SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 2105e02119d5SChris Mason log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2106e02119d5SChris Mason 2107e02119d5SChris Mason log_tree_root->node = read_tree_block(tree_root, bytenr, 210884234f3aSYan Zheng blocksize, 210984234f3aSYan Zheng generation + 1); 2110e02119d5SChris Mason ret = btrfs_recover_log_trees(log_tree_root); 2111e02119d5SChris Mason BUG_ON(ret); 2112e556ce2cSYan Zheng 2113e556ce2cSYan Zheng if (sb->s_flags & MS_RDONLY) { 2114e556ce2cSYan Zheng ret = btrfs_commit_super(tree_root); 2115e556ce2cSYan Zheng BUG_ON(ret); 2116e556ce2cSYan Zheng } 2117e02119d5SChris Mason } 21181a40e23bSZheng Yan 211976dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 212076dda93cSYan, Zheng BUG_ON(ret); 212176dda93cSYan, Zheng 21227c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 2123d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 2124d68fc57bSYan, Zheng BUG_ON(ret); 2125d68fc57bSYan, Zheng 21265d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 2127d7ce5843SMiao Xie if (ret < 0) { 2128d7ce5843SMiao Xie printk(KERN_WARNING 2129d7ce5843SMiao Xie "btrfs: failed to recover relocation\n"); 2130d7ce5843SMiao Xie err = -EINVAL; 2131d7ce5843SMiao Xie goto fail_trans_kthread; 2132d7ce5843SMiao Xie } 21337c2ca468SChris Mason } 21341a40e23bSZheng Yan 21353de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 21363de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 21373de4586cSChris Mason location.offset = (u64)-1; 21383de4586cSChris Mason 21393de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 21403de4586cSChris Mason if (!fs_info->fs_root) 21417c2ca468SChris Mason goto fail_trans_kthread; 21423140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 21433140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 21443140c9a3SDan Carpenter goto fail_trans_kthread; 21453140c9a3SDan Carpenter } 2146c289811cSChris Mason 2147e3acc2a6SJosef Bacik if (!(sb->s_flags & MS_RDONLY)) { 2148e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 214966b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_info->fs_root); 215066b4ffd1SJosef Bacik if (!err) 215166b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_info->tree_root); 2152e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 215366b4ffd1SJosef Bacik if (err) { 215466b4ffd1SJosef Bacik close_ctree(tree_root); 215566b4ffd1SJosef Bacik return ERR_PTR(err); 215666b4ffd1SJosef Bacik } 2157e3acc2a6SJosef Bacik } 2158e3acc2a6SJosef Bacik 21590f7d52f4SChris Mason return tree_root; 216039279cc3SChris Mason 21617c2ca468SChris Mason fail_trans_kthread: 21627c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 21633f157a2fSChris Mason fail_cleaner: 2164a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 21657c2ca468SChris Mason 21667c2ca468SChris Mason /* 21677c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 21687c2ca468SChris Mason * kthreads 21697c2ca468SChris Mason */ 21707c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 21717c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 21727c2ca468SChris Mason 21731b1d1f66SJosef Bacik fail_block_groups: 21741b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 2175d20f7043SChris Mason free_extent_buffer(csum_root->node); 21765d4f98a2SYan Zheng free_extent_buffer(csum_root->commit_root); 21775d4f98a2SYan Zheng fail_dev_root: 21785d4f98a2SYan Zheng free_extent_buffer(dev_root->node); 21795d4f98a2SYan Zheng free_extent_buffer(dev_root->commit_root); 21800b86a832SChris Mason fail_extent_root: 21810b86a832SChris Mason free_extent_buffer(extent_root->node); 21825d4f98a2SYan Zheng free_extent_buffer(extent_root->commit_root); 218339279cc3SChris Mason fail_tree_root: 21845f39d397SChris Mason free_extent_buffer(tree_root->node); 21855d4f98a2SYan Zheng free_extent_buffer(tree_root->commit_root); 21862b82032cSYan Zheng fail_chunk_root: 21872b82032cSYan Zheng free_extent_buffer(chunk_root->node); 21885d4f98a2SYan Zheng free_extent_buffer(chunk_root->commit_root); 218939279cc3SChris Mason fail_sb_buffer: 219061d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 2191247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 2192771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 21938b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 21948b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 2195d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 2196cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2197e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 21980cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 21991cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 22004543df7eSChris Mason fail_iput: 22017c2ca468SChris Mason invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 22024543df7eSChris Mason iput(fs_info->btree_inode); 22037e662854SQinghuang Feng 2204dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 220584eed90fSChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 2206ad081f14SJens Axboe fail_bdi: 22077e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 220876dda93cSYan, Zheng fail_srcu: 220976dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 22107e662854SQinghuang Feng fail: 221139279cc3SChris Mason kfree(extent_root); 221239279cc3SChris Mason kfree(tree_root); 221339279cc3SChris Mason kfree(fs_info); 221483afeac4SJim Meyering kfree(chunk_root); 221583afeac4SJim Meyering kfree(dev_root); 2216d20f7043SChris Mason kfree(csum_root); 221739279cc3SChris Mason return ERR_PTR(err); 2218eb60ceacSChris Mason } 2219eb60ceacSChris Mason 2220f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 2221f2984462SChris Mason { 2222f2984462SChris Mason char b[BDEVNAME_SIZE]; 2223f2984462SChris Mason 2224f2984462SChris Mason if (uptodate) { 2225f2984462SChris Mason set_buffer_uptodate(bh); 2226f2984462SChris Mason } else { 2227c3b9a62cSChristoph Hellwig if (printk_ratelimit()) { 2228f2984462SChris Mason printk(KERN_WARNING "lost page write due to " 2229f2984462SChris Mason "I/O error on %s\n", 2230f2984462SChris Mason bdevname(bh->b_bdev, b)); 2231f2984462SChris Mason } 22321259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 22331259ab75SChris Mason * our own ways of dealing with the IO errors 22341259ab75SChris Mason */ 2235f2984462SChris Mason clear_buffer_uptodate(bh); 2236f2984462SChris Mason } 2237f2984462SChris Mason unlock_buffer(bh); 2238f2984462SChris Mason put_bh(bh); 2239f2984462SChris Mason } 2240f2984462SChris Mason 2241a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 2242a512bbf8SYan Zheng { 2243a512bbf8SYan Zheng struct buffer_head *bh; 2244a512bbf8SYan Zheng struct buffer_head *latest = NULL; 2245a512bbf8SYan Zheng struct btrfs_super_block *super; 2246a512bbf8SYan Zheng int i; 2247a512bbf8SYan Zheng u64 transid = 0; 2248a512bbf8SYan Zheng u64 bytenr; 2249a512bbf8SYan Zheng 2250a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 2251a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 2252a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 2253a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 2254a512bbf8SYan Zheng */ 2255a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 2256a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2257a512bbf8SYan Zheng if (bytenr + 4096 >= i_size_read(bdev->bd_inode)) 2258a512bbf8SYan Zheng break; 2259a512bbf8SYan Zheng bh = __bread(bdev, bytenr / 4096, 4096); 2260a512bbf8SYan Zheng if (!bh) 2261a512bbf8SYan Zheng continue; 2262a512bbf8SYan Zheng 2263a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 2264a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 2265a512bbf8SYan Zheng strncmp((char *)(&super->magic), BTRFS_MAGIC, 2266a512bbf8SYan Zheng sizeof(super->magic))) { 2267a512bbf8SYan Zheng brelse(bh); 2268a512bbf8SYan Zheng continue; 2269a512bbf8SYan Zheng } 2270a512bbf8SYan Zheng 2271a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 2272a512bbf8SYan Zheng brelse(latest); 2273a512bbf8SYan Zheng latest = bh; 2274a512bbf8SYan Zheng transid = btrfs_super_generation(super); 2275a512bbf8SYan Zheng } else { 2276a512bbf8SYan Zheng brelse(bh); 2277a512bbf8SYan Zheng } 2278a512bbf8SYan Zheng } 2279a512bbf8SYan Zheng return latest; 2280a512bbf8SYan Zheng } 2281a512bbf8SYan Zheng 22824eedeb75SHisashi Hifumi /* 22834eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 22844eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 22854eedeb75SHisashi Hifumi * we write are pinned. 22864eedeb75SHisashi Hifumi * 22874eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 22884eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 22894eedeb75SHisashi Hifumi * many supers on this one device should be written. 22904eedeb75SHisashi Hifumi * 22914eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 22924eedeb75SHisashi Hifumi */ 2293a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 2294a512bbf8SYan Zheng struct btrfs_super_block *sb, 2295a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 2296a512bbf8SYan Zheng { 2297a512bbf8SYan Zheng struct buffer_head *bh; 2298a512bbf8SYan Zheng int i; 2299a512bbf8SYan Zheng int ret; 2300a512bbf8SYan Zheng int errors = 0; 2301a512bbf8SYan Zheng u32 crc; 2302a512bbf8SYan Zheng u64 bytenr; 2303a512bbf8SYan Zheng int last_barrier = 0; 2304a512bbf8SYan Zheng 2305a512bbf8SYan Zheng if (max_mirrors == 0) 2306a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 2307a512bbf8SYan Zheng 2308a512bbf8SYan Zheng /* make sure only the last submit_bh does a barrier */ 2309a512bbf8SYan Zheng if (do_barriers) { 2310a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 2311a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2312a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= 2313a512bbf8SYan Zheng device->total_bytes) 2314a512bbf8SYan Zheng break; 2315a512bbf8SYan Zheng last_barrier = i; 2316a512bbf8SYan Zheng } 2317a512bbf8SYan Zheng } 2318a512bbf8SYan Zheng 2319a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 2320a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 2321a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 2322a512bbf8SYan Zheng break; 2323a512bbf8SYan Zheng 2324a512bbf8SYan Zheng if (wait) { 2325a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 2326a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2327a512bbf8SYan Zheng BUG_ON(!bh); 2328a512bbf8SYan Zheng wait_on_buffer(bh); 23294eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 23304eedeb75SHisashi Hifumi errors++; 23314eedeb75SHisashi Hifumi 23324eedeb75SHisashi Hifumi /* drop our reference */ 23334eedeb75SHisashi Hifumi brelse(bh); 23344eedeb75SHisashi Hifumi 23354eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 2336a512bbf8SYan Zheng brelse(bh); 2337a512bbf8SYan Zheng continue; 2338a512bbf8SYan Zheng } else { 2339a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 2340a512bbf8SYan Zheng 2341a512bbf8SYan Zheng crc = ~(u32)0; 2342a512bbf8SYan Zheng crc = btrfs_csum_data(NULL, (char *)sb + 2343a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 2344a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 2345a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 2346a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 2347a512bbf8SYan Zheng 23484eedeb75SHisashi Hifumi /* 23494eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 23504eedeb75SHisashi Hifumi * caller 23514eedeb75SHisashi Hifumi */ 2352a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 2353a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 2354a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 2355a512bbf8SYan Zheng 23564eedeb75SHisashi Hifumi /* one reference for submit_bh */ 2357a512bbf8SYan Zheng get_bh(bh); 23584eedeb75SHisashi Hifumi 23594eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 2360a512bbf8SYan Zheng lock_buffer(bh); 2361a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 2362a512bbf8SYan Zheng } 2363a512bbf8SYan Zheng 2364c3b9a62cSChristoph Hellwig if (i == last_barrier && do_barriers) 2365c3b9a62cSChristoph Hellwig ret = submit_bh(WRITE_FLUSH_FUA, bh); 2366c3b9a62cSChristoph Hellwig else 2367ffbd517dSChris Mason ret = submit_bh(WRITE_SYNC, bh); 2368a512bbf8SYan Zheng 23694eedeb75SHisashi Hifumi if (ret) 2370a512bbf8SYan Zheng errors++; 2371a512bbf8SYan Zheng } 2372a512bbf8SYan Zheng return errors < i ? 0 : -1; 2373a512bbf8SYan Zheng } 2374a512bbf8SYan Zheng 2375a512bbf8SYan Zheng int write_all_supers(struct btrfs_root *root, int max_mirrors) 2376f2984462SChris Mason { 2377e5e9a520SChris Mason struct list_head *head; 2378f2984462SChris Mason struct btrfs_device *dev; 2379a061fc8dSChris Mason struct btrfs_super_block *sb; 2380f2984462SChris Mason struct btrfs_dev_item *dev_item; 2381f2984462SChris Mason int ret; 2382f2984462SChris Mason int do_barriers; 2383a236aed1SChris Mason int max_errors; 2384a236aed1SChris Mason int total_errors = 0; 2385a061fc8dSChris Mason u64 flags; 2386f2984462SChris Mason 2387a236aed1SChris Mason max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1; 2388f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 2389f2984462SChris Mason 2390a061fc8dSChris Mason sb = &root->fs_info->super_for_commit; 2391a061fc8dSChris Mason dev_item = &sb->dev_item; 2392e5e9a520SChris Mason 2393e5e9a520SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 2394e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 2395c6e30871SQinghuang Feng list_for_each_entry(dev, head, dev_list) { 2396dfe25020SChris Mason if (!dev->bdev) { 2397dfe25020SChris Mason total_errors++; 2398dfe25020SChris Mason continue; 2399dfe25020SChris Mason } 24002b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 2401dfe25020SChris Mason continue; 2402dfe25020SChris Mason 24032b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 2404a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 2405a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 2406a061fc8dSChris Mason btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 2407a061fc8dSChris Mason btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 2408a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 2409a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 2410a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 2411a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 24122b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 2413a512bbf8SYan Zheng 2414a061fc8dSChris Mason flags = btrfs_super_flags(sb); 2415a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 2416f2984462SChris Mason 2417a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 2418a236aed1SChris Mason if (ret) 2419a236aed1SChris Mason total_errors++; 2420f2984462SChris Mason } 2421a236aed1SChris Mason if (total_errors > max_errors) { 2422d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 2423d397712bSChris Mason total_errors); 2424a236aed1SChris Mason BUG(); 2425a236aed1SChris Mason } 2426f2984462SChris Mason 2427a512bbf8SYan Zheng total_errors = 0; 2428c6e30871SQinghuang Feng list_for_each_entry(dev, head, dev_list) { 2429dfe25020SChris Mason if (!dev->bdev) 2430dfe25020SChris Mason continue; 24312b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 2432dfe25020SChris Mason continue; 2433dfe25020SChris Mason 2434a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 2435a512bbf8SYan Zheng if (ret) 24361259ab75SChris Mason total_errors++; 2437f2984462SChris Mason } 2438e5e9a520SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 2439a236aed1SChris Mason if (total_errors > max_errors) { 2440d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 2441d397712bSChris Mason total_errors); 2442a236aed1SChris Mason BUG(); 2443a236aed1SChris Mason } 2444f2984462SChris Mason return 0; 2445f2984462SChris Mason } 2446f2984462SChris Mason 2447a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 2448a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 2449cfaa7295SChris Mason { 2450e66f709bSChris Mason int ret; 24512c90e5d6SChris Mason 2452a512bbf8SYan Zheng ret = write_all_supers(root, max_mirrors); 24535f39d397SChris Mason return ret; 2454cfaa7295SChris Mason } 2455cfaa7295SChris Mason 24565eda7b5eSChris Mason int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root) 24572619ba1fSChris Mason { 24584df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 24592619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 24602619ba1fSChris Mason (unsigned long)root->root_key.objectid); 24614df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 246276dda93cSYan, Zheng 246376dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 246476dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 246576dda93cSYan, Zheng 24664df27c4dSYan, Zheng free_fs_root(root); 24674df27c4dSYan, Zheng return 0; 24684df27c4dSYan, Zheng } 24694df27c4dSYan, Zheng 24704df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 24714df27c4dSYan, Zheng { 24724df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 24733394e160SChris Mason if (root->anon_super.s_dev) { 24743394e160SChris Mason down_write(&root->anon_super.s_umount); 24753394e160SChris Mason kill_anon_super(&root->anon_super); 24763394e160SChris Mason } 24775f39d397SChris Mason free_extent_buffer(root->node); 24785f39d397SChris Mason free_extent_buffer(root->commit_root); 247958176a96SJosef Bacik kfree(root->name); 24802619ba1fSChris Mason kfree(root); 24812619ba1fSChris Mason } 24822619ba1fSChris Mason 248335b7e476SChris Mason static int del_fs_roots(struct btrfs_fs_info *fs_info) 24840f7d52f4SChris Mason { 24850f7d52f4SChris Mason int ret; 24860f7d52f4SChris Mason struct btrfs_root *gang[8]; 24870f7d52f4SChris Mason int i; 24880f7d52f4SChris Mason 248976dda93cSYan, Zheng while (!list_empty(&fs_info->dead_roots)) { 249076dda93cSYan, Zheng gang[0] = list_entry(fs_info->dead_roots.next, 249176dda93cSYan, Zheng struct btrfs_root, root_list); 249276dda93cSYan, Zheng list_del(&gang[0]->root_list); 249376dda93cSYan, Zheng 249476dda93cSYan, Zheng if (gang[0]->in_radix) { 249576dda93cSYan, Zheng btrfs_free_fs_root(fs_info, gang[0]); 249676dda93cSYan, Zheng } else { 249776dda93cSYan, Zheng free_extent_buffer(gang[0]->node); 249876dda93cSYan, Zheng free_extent_buffer(gang[0]->commit_root); 249976dda93cSYan, Zheng kfree(gang[0]); 250076dda93cSYan, Zheng } 250176dda93cSYan, Zheng } 250276dda93cSYan, Zheng 25030f7d52f4SChris Mason while (1) { 25040f7d52f4SChris Mason ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 25050f7d52f4SChris Mason (void **)gang, 0, 25060f7d52f4SChris Mason ARRAY_SIZE(gang)); 25070f7d52f4SChris Mason if (!ret) 25080f7d52f4SChris Mason break; 25092619ba1fSChris Mason for (i = 0; i < ret; i++) 25105eda7b5eSChris Mason btrfs_free_fs_root(fs_info, gang[i]); 25110f7d52f4SChris Mason } 25120f7d52f4SChris Mason return 0; 25130f7d52f4SChris Mason } 2514b4100d64SChris Mason 2515c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 2516c146afadSYan Zheng { 2517c146afadSYan Zheng u64 root_objectid = 0; 2518c146afadSYan Zheng struct btrfs_root *gang[8]; 2519c146afadSYan Zheng int i; 2520c146afadSYan Zheng int ret; 2521c146afadSYan Zheng 2522c146afadSYan Zheng while (1) { 2523c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2524c146afadSYan Zheng (void **)gang, root_objectid, 2525c146afadSYan Zheng ARRAY_SIZE(gang)); 2526c146afadSYan Zheng if (!ret) 2527c146afadSYan Zheng break; 25285d4f98a2SYan Zheng 25295d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 2530c146afadSYan Zheng for (i = 0; i < ret; i++) { 253166b4ffd1SJosef Bacik int err; 253266b4ffd1SJosef Bacik 2533c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 253466b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 253566b4ffd1SJosef Bacik if (err) 253666b4ffd1SJosef Bacik return err; 2537c146afadSYan Zheng } 2538c146afadSYan Zheng root_objectid++; 2539c146afadSYan Zheng } 2540c146afadSYan Zheng return 0; 2541c146afadSYan Zheng } 2542c146afadSYan Zheng 2543c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 2544c146afadSYan Zheng { 2545c146afadSYan Zheng struct btrfs_trans_handle *trans; 2546c146afadSYan Zheng int ret; 2547c146afadSYan Zheng 2548c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 254924bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 2550c146afadSYan Zheng btrfs_clean_old_snapshots(root); 2551c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 2552c71bf099SYan, Zheng 2553c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 2554c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 2555c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 2556c71bf099SYan, Zheng 2557a22285a6SYan, Zheng trans = btrfs_join_transaction(root, 1); 25583612b495STsutomu Itoh if (IS_ERR(trans)) 25593612b495STsutomu Itoh return PTR_ERR(trans); 2560c146afadSYan Zheng ret = btrfs_commit_transaction(trans, root); 2561c146afadSYan Zheng BUG_ON(ret); 2562c146afadSYan Zheng /* run commit again to drop the original snapshot */ 2563a22285a6SYan, Zheng trans = btrfs_join_transaction(root, 1); 25643612b495STsutomu Itoh if (IS_ERR(trans)) 25653612b495STsutomu Itoh return PTR_ERR(trans); 2566c146afadSYan Zheng btrfs_commit_transaction(trans, root); 2567c146afadSYan Zheng ret = btrfs_write_and_wait_transaction(NULL, root); 2568c146afadSYan Zheng BUG_ON(ret); 2569c146afadSYan Zheng 2570a512bbf8SYan Zheng ret = write_ctree_super(NULL, root, 0); 2571c146afadSYan Zheng return ret; 2572c146afadSYan Zheng } 2573c146afadSYan Zheng 2574e20d96d6SChris Mason int close_ctree(struct btrfs_root *root) 2575eb60ceacSChris Mason { 25760f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 2577c146afadSYan Zheng int ret; 2578e089f05cSChris Mason 2579facda1e7SChris Mason fs_info->closing = 1; 2580a2135011SChris Mason smp_mb(); 2581a2135011SChris Mason 25820af3d00bSJosef Bacik btrfs_put_block_group_cache(fs_info); 2583acce952bSliubo 2584acce952bSliubo /* 2585acce952bSliubo * Here come 2 situations when btrfs is broken to flip readonly: 2586acce952bSliubo * 2587acce952bSliubo * 1. when btrfs flips readonly somewhere else before 2588acce952bSliubo * btrfs_commit_super, sb->s_flags has MS_RDONLY flag, 2589acce952bSliubo * and btrfs will skip to write sb directly to keep 2590acce952bSliubo * ERROR state on disk. 2591acce952bSliubo * 2592acce952bSliubo * 2. when btrfs flips readonly just in btrfs_commit_super, 2593acce952bSliubo * and in such case, btrfs cannnot write sb via btrfs_commit_super, 2594acce952bSliubo * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag, 2595acce952bSliubo * btrfs will cleanup all FS resources first and write sb then. 2596acce952bSliubo */ 2597c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 2598c146afadSYan Zheng ret = btrfs_commit_super(root); 2599d397712bSChris Mason if (ret) 2600d397712bSChris Mason printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 2601c146afadSYan Zheng } 2602ed2ff2cbSChris Mason 2603acce952bSliubo if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { 2604acce952bSliubo ret = btrfs_error_commit_super(root); 2605acce952bSliubo if (ret) 2606acce952bSliubo printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 2607acce952bSliubo } 2608acce952bSliubo 26098929ecfaSYan, Zheng kthread_stop(root->fs_info->transaction_kthread); 26108929ecfaSYan, Zheng kthread_stop(root->fs_info->cleaner_kthread); 26118929ecfaSYan, Zheng 2612f25784b3SYan Zheng fs_info->closing = 2; 2613f25784b3SYan Zheng smp_mb(); 2614f25784b3SYan Zheng 2615b0c68f8bSChris Mason if (fs_info->delalloc_bytes) { 2616d397712bSChris Mason printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n", 261721380931SJoel Becker (unsigned long long)fs_info->delalloc_bytes); 2618b0c68f8bSChris Mason } 261931153d81SYan Zheng if (fs_info->total_ref_cache_size) { 2620d397712bSChris Mason printk(KERN_INFO "btrfs: at umount reference cache size %llu\n", 2621d397712bSChris Mason (unsigned long long)fs_info->total_ref_cache_size); 262231153d81SYan Zheng } 262331153d81SYan Zheng 26245f39d397SChris Mason free_extent_buffer(fs_info->extent_root->node); 26255d4f98a2SYan Zheng free_extent_buffer(fs_info->extent_root->commit_root); 26265f39d397SChris Mason free_extent_buffer(fs_info->tree_root->node); 26275d4f98a2SYan Zheng free_extent_buffer(fs_info->tree_root->commit_root); 26280b86a832SChris Mason free_extent_buffer(root->fs_info->chunk_root->node); 26295d4f98a2SYan Zheng free_extent_buffer(root->fs_info->chunk_root->commit_root); 26300b86a832SChris Mason free_extent_buffer(root->fs_info->dev_root->node); 26315d4f98a2SYan Zheng free_extent_buffer(root->fs_info->dev_root->commit_root); 2632d20f7043SChris Mason free_extent_buffer(root->fs_info->csum_root->node); 26335d4f98a2SYan Zheng free_extent_buffer(root->fs_info->csum_root->commit_root); 2634d20f7043SChris Mason 26359078a3e1SChris Mason btrfs_free_block_groups(root->fs_info); 2636c146afadSYan Zheng 26370f7d52f4SChris Mason del_fs_roots(fs_info); 2638d10c5f31SChris Mason 2639c146afadSYan Zheng iput(fs_info->btree_inode); 26409ad6b7bcSChris Mason 264161d92c32SChris Mason btrfs_stop_workers(&fs_info->generic_worker); 2642247e743cSChris Mason btrfs_stop_workers(&fs_info->fixup_workers); 2643771ed689SChris Mason btrfs_stop_workers(&fs_info->delalloc_workers); 26448b712842SChris Mason btrfs_stop_workers(&fs_info->workers); 26458b712842SChris Mason btrfs_stop_workers(&fs_info->endio_workers); 2646d20f7043SChris Mason btrfs_stop_workers(&fs_info->endio_meta_workers); 2647cad321adSChris Mason btrfs_stop_workers(&fs_info->endio_meta_write_workers); 2648e6dcd2dcSChris Mason btrfs_stop_workers(&fs_info->endio_write_workers); 26490cb59c99SJosef Bacik btrfs_stop_workers(&fs_info->endio_freespace_worker); 26501cc127b5SChris Mason btrfs_stop_workers(&fs_info->submit_workers); 2651d6bfde87SChris Mason 2652dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 26530b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 2654b248a415SChris Mason 265504160088SChris Mason bdi_destroy(&fs_info->bdi); 265676dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 26570b86a832SChris Mason 26580f7d52f4SChris Mason kfree(fs_info->extent_root); 26590f7d52f4SChris Mason kfree(fs_info->tree_root); 26600b86a832SChris Mason kfree(fs_info->chunk_root); 26610b86a832SChris Mason kfree(fs_info->dev_root); 2662d20f7043SChris Mason kfree(fs_info->csum_root); 266383a4d548SLi Zefan kfree(fs_info); 266483a4d548SLi Zefan 2665eb60ceacSChris Mason return 0; 2666eb60ceacSChris Mason } 2667eb60ceacSChris Mason 26681259ab75SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid) 2669ccd467d6SChris Mason { 26701259ab75SChris Mason int ret; 2671810191ffSChris Mason struct inode *btree_inode = buf->first_page->mapping->host; 26721259ab75SChris Mason 26732ac55d41SJosef Bacik ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf, 26742ac55d41SJosef Bacik NULL); 26751259ab75SChris Mason if (!ret) 26761259ab75SChris Mason return ret; 26771259ab75SChris Mason 26781259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 26791259ab75SChris Mason parent_transid); 26801259ab75SChris Mason return !ret; 26815f39d397SChris Mason } 26826702ed49SChris Mason 26835f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 26845f39d397SChris Mason { 2685810191ffSChris Mason struct inode *btree_inode = buf->first_page->mapping->host; 2686d1310b2eSChris Mason return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, 26875f39d397SChris Mason buf); 26885f39d397SChris Mason } 26895f39d397SChris Mason 26905f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 26915f39d397SChris Mason { 2692810191ffSChris Mason struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 26935f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 26945f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 2695b9473439SChris Mason int was_dirty; 2696b4ce94deSChris Mason 2697b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 2698ccd467d6SChris Mason if (transid != root->fs_info->generation) { 2699d397712bSChris Mason printk(KERN_CRIT "btrfs transid mismatch buffer %llu, " 2700d397712bSChris Mason "found %llu running %llu\n", 2701db94535dSChris Mason (unsigned long long)buf->start, 2702d397712bSChris Mason (unsigned long long)transid, 2703d397712bSChris Mason (unsigned long long)root->fs_info->generation); 2704ccd467d6SChris Mason WARN_ON(1); 2705ccd467d6SChris Mason } 2706b9473439SChris Mason was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 2707b9473439SChris Mason buf); 2708b9473439SChris Mason if (!was_dirty) { 2709b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 2710b9473439SChris Mason root->fs_info->dirty_metadata_bytes += buf->len; 2711b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 2712b9473439SChris Mason } 2713eb60ceacSChris Mason } 2714eb60ceacSChris Mason 2715d3c2fdcfSChris Mason void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) 271635b7e476SChris Mason { 2717188de649SChris Mason /* 2718188de649SChris Mason * looks as though older kernels can get into trouble with 2719188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 2720188de649SChris Mason */ 2721d6bfde87SChris Mason u64 num_dirty; 2722771ed689SChris Mason unsigned long thresh = 32 * 1024 * 1024; 2723d6bfde87SChris Mason 27246933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 2725d6bfde87SChris Mason return; 2726d6bfde87SChris Mason 2727585ad2c3SChris Mason num_dirty = root->fs_info->dirty_metadata_bytes; 2728585ad2c3SChris Mason 2729d6bfde87SChris Mason if (num_dirty > thresh) { 2730d3c2fdcfSChris Mason balance_dirty_pages_ratelimited_nr( 2731304fced6SChris Mason root->fs_info->btree_inode->i_mapping, 1); 273235b7e476SChris Mason } 2733188de649SChris Mason return; 2734d6bfde87SChris Mason } 27356b80053dSChris Mason 2736ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 27376b80053dSChris Mason { 2738810191ffSChris Mason struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 2739ce9adaa5SChris Mason int ret; 2740ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 2741d397712bSChris Mason if (ret == 0) 2742b4ce94deSChris Mason set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags); 2743ce9adaa5SChris Mason return ret; 27446b80053dSChris Mason } 27450da5468fSChris Mason 27464bef0848SChris Mason int btree_lock_page_hook(struct page *page) 27474bef0848SChris Mason { 27484bef0848SChris Mason struct inode *inode = page->mapping->host; 2749b9473439SChris Mason struct btrfs_root *root = BTRFS_I(inode)->root; 27504bef0848SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 27514bef0848SChris Mason struct extent_buffer *eb; 27524bef0848SChris Mason unsigned long len; 27534bef0848SChris Mason u64 bytenr = page_offset(page); 27544bef0848SChris Mason 27554bef0848SChris Mason if (page->private == EXTENT_PAGE_PRIVATE) 27564bef0848SChris Mason goto out; 27574bef0848SChris Mason 27584bef0848SChris Mason len = page->private >> 2; 27594bef0848SChris Mason eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS); 27604bef0848SChris Mason if (!eb) 27614bef0848SChris Mason goto out; 27624bef0848SChris Mason 27634bef0848SChris Mason btrfs_tree_lock(eb); 27644bef0848SChris Mason btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); 2765b9473439SChris Mason 2766b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { 2767b9473439SChris Mason spin_lock(&root->fs_info->delalloc_lock); 2768b9473439SChris Mason if (root->fs_info->dirty_metadata_bytes >= eb->len) 2769b9473439SChris Mason root->fs_info->dirty_metadata_bytes -= eb->len; 2770b9473439SChris Mason else 2771b9473439SChris Mason WARN_ON(1); 2772b9473439SChris Mason spin_unlock(&root->fs_info->delalloc_lock); 2773b9473439SChris Mason } 2774b9473439SChris Mason 27754bef0848SChris Mason btrfs_tree_unlock(eb); 27764bef0848SChris Mason free_extent_buffer(eb); 27774bef0848SChris Mason out: 27784bef0848SChris Mason lock_page(page); 27794bef0848SChris Mason return 0; 27804bef0848SChris Mason } 27814bef0848SChris Mason 2782acce952bSliubo static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 2783acce952bSliubo int read_only) 2784acce952bSliubo { 2785acce952bSliubo if (read_only) 2786acce952bSliubo return; 2787acce952bSliubo 2788acce952bSliubo if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 2789acce952bSliubo printk(KERN_WARNING "warning: mount fs with errors, " 2790acce952bSliubo "running btrfsck is recommended\n"); 2791acce952bSliubo } 2792acce952bSliubo 2793acce952bSliubo int btrfs_error_commit_super(struct btrfs_root *root) 2794acce952bSliubo { 2795acce952bSliubo int ret; 2796acce952bSliubo 2797acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 2798acce952bSliubo btrfs_run_delayed_iputs(root); 2799acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 2800acce952bSliubo 2801acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 2802acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 2803acce952bSliubo 2804acce952bSliubo /* cleanup FS via transaction */ 2805acce952bSliubo btrfs_cleanup_transaction(root); 2806acce952bSliubo 2807acce952bSliubo ret = write_ctree_super(NULL, root, 0); 2808acce952bSliubo 2809acce952bSliubo return ret; 2810acce952bSliubo } 2811acce952bSliubo 2812acce952bSliubo static int btrfs_destroy_ordered_operations(struct btrfs_root *root) 2813acce952bSliubo { 2814acce952bSliubo struct btrfs_inode *btrfs_inode; 2815acce952bSliubo struct list_head splice; 2816acce952bSliubo 2817acce952bSliubo INIT_LIST_HEAD(&splice); 2818acce952bSliubo 2819acce952bSliubo mutex_lock(&root->fs_info->ordered_operations_mutex); 2820acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2821acce952bSliubo 2822acce952bSliubo list_splice_init(&root->fs_info->ordered_operations, &splice); 2823acce952bSliubo while (!list_empty(&splice)) { 2824acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 2825acce952bSliubo ordered_operations); 2826acce952bSliubo 2827acce952bSliubo list_del_init(&btrfs_inode->ordered_operations); 2828acce952bSliubo 2829acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 2830acce952bSliubo } 2831acce952bSliubo 2832acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2833acce952bSliubo mutex_unlock(&root->fs_info->ordered_operations_mutex); 2834acce952bSliubo 2835acce952bSliubo return 0; 2836acce952bSliubo } 2837acce952bSliubo 2838acce952bSliubo static int btrfs_destroy_ordered_extents(struct btrfs_root *root) 2839acce952bSliubo { 2840acce952bSliubo struct list_head splice; 2841acce952bSliubo struct btrfs_ordered_extent *ordered; 2842acce952bSliubo struct inode *inode; 2843acce952bSliubo 2844acce952bSliubo INIT_LIST_HEAD(&splice); 2845acce952bSliubo 2846acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2847acce952bSliubo 2848acce952bSliubo list_splice_init(&root->fs_info->ordered_extents, &splice); 2849acce952bSliubo while (!list_empty(&splice)) { 2850acce952bSliubo ordered = list_entry(splice.next, struct btrfs_ordered_extent, 2851acce952bSliubo root_extent_list); 2852acce952bSliubo 2853acce952bSliubo list_del_init(&ordered->root_extent_list); 2854acce952bSliubo atomic_inc(&ordered->refs); 2855acce952bSliubo 2856acce952bSliubo /* the inode may be getting freed (in sys_unlink path). */ 2857acce952bSliubo inode = igrab(ordered->inode); 2858acce952bSliubo 2859acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2860acce952bSliubo if (inode) 2861acce952bSliubo iput(inode); 2862acce952bSliubo 2863acce952bSliubo atomic_set(&ordered->refs, 1); 2864acce952bSliubo btrfs_put_ordered_extent(ordered); 2865acce952bSliubo 2866acce952bSliubo spin_lock(&root->fs_info->ordered_extent_lock); 2867acce952bSliubo } 2868acce952bSliubo 2869acce952bSliubo spin_unlock(&root->fs_info->ordered_extent_lock); 2870acce952bSliubo 2871acce952bSliubo return 0; 2872acce952bSliubo } 2873acce952bSliubo 2874acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 2875acce952bSliubo struct btrfs_root *root) 2876acce952bSliubo { 2877acce952bSliubo struct rb_node *node; 2878acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 2879acce952bSliubo struct btrfs_delayed_ref_node *ref; 2880acce952bSliubo int ret = 0; 2881acce952bSliubo 2882acce952bSliubo delayed_refs = &trans->delayed_refs; 2883acce952bSliubo 2884acce952bSliubo spin_lock(&delayed_refs->lock); 2885acce952bSliubo if (delayed_refs->num_entries == 0) { 2886acce952bSliubo printk(KERN_INFO "delayed_refs has NO entry\n"); 2887acce952bSliubo return ret; 2888acce952bSliubo } 2889acce952bSliubo 2890acce952bSliubo node = rb_first(&delayed_refs->root); 2891acce952bSliubo while (node) { 2892acce952bSliubo ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); 2893acce952bSliubo node = rb_next(node); 2894acce952bSliubo 2895acce952bSliubo ref->in_tree = 0; 2896acce952bSliubo rb_erase(&ref->rb_node, &delayed_refs->root); 2897acce952bSliubo delayed_refs->num_entries--; 2898acce952bSliubo 2899acce952bSliubo atomic_set(&ref->refs, 1); 2900acce952bSliubo if (btrfs_delayed_ref_is_head(ref)) { 2901acce952bSliubo struct btrfs_delayed_ref_head *head; 2902acce952bSliubo 2903acce952bSliubo head = btrfs_delayed_node_to_head(ref); 2904acce952bSliubo mutex_lock(&head->mutex); 2905acce952bSliubo kfree(head->extent_op); 2906acce952bSliubo delayed_refs->num_heads--; 2907acce952bSliubo if (list_empty(&head->cluster)) 2908acce952bSliubo delayed_refs->num_heads_ready--; 2909acce952bSliubo list_del_init(&head->cluster); 2910acce952bSliubo mutex_unlock(&head->mutex); 2911acce952bSliubo } 2912acce952bSliubo 2913acce952bSliubo spin_unlock(&delayed_refs->lock); 2914acce952bSliubo btrfs_put_delayed_ref(ref); 2915acce952bSliubo 2916acce952bSliubo cond_resched(); 2917acce952bSliubo spin_lock(&delayed_refs->lock); 2918acce952bSliubo } 2919acce952bSliubo 2920acce952bSliubo spin_unlock(&delayed_refs->lock); 2921acce952bSliubo 2922acce952bSliubo return ret; 2923acce952bSliubo } 2924acce952bSliubo 2925acce952bSliubo static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t) 2926acce952bSliubo { 2927acce952bSliubo struct btrfs_pending_snapshot *snapshot; 2928acce952bSliubo struct list_head splice; 2929acce952bSliubo 2930acce952bSliubo INIT_LIST_HEAD(&splice); 2931acce952bSliubo 2932acce952bSliubo list_splice_init(&t->pending_snapshots, &splice); 2933acce952bSliubo 2934acce952bSliubo while (!list_empty(&splice)) { 2935acce952bSliubo snapshot = list_entry(splice.next, 2936acce952bSliubo struct btrfs_pending_snapshot, 2937acce952bSliubo list); 2938acce952bSliubo 2939acce952bSliubo list_del_init(&snapshot->list); 2940acce952bSliubo 2941acce952bSliubo kfree(snapshot); 2942acce952bSliubo } 2943acce952bSliubo 2944acce952bSliubo return 0; 2945acce952bSliubo } 2946acce952bSliubo 2947acce952bSliubo static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 2948acce952bSliubo { 2949acce952bSliubo struct btrfs_inode *btrfs_inode; 2950acce952bSliubo struct list_head splice; 2951acce952bSliubo 2952acce952bSliubo INIT_LIST_HEAD(&splice); 2953acce952bSliubo 2954acce952bSliubo list_splice_init(&root->fs_info->delalloc_inodes, &splice); 2955acce952bSliubo 2956acce952bSliubo spin_lock(&root->fs_info->delalloc_lock); 2957acce952bSliubo 2958acce952bSliubo while (!list_empty(&splice)) { 2959acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 2960acce952bSliubo delalloc_inodes); 2961acce952bSliubo 2962acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 2963acce952bSliubo 2964acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 2965acce952bSliubo } 2966acce952bSliubo 2967acce952bSliubo spin_unlock(&root->fs_info->delalloc_lock); 2968acce952bSliubo 2969acce952bSliubo return 0; 2970acce952bSliubo } 2971acce952bSliubo 2972acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 2973acce952bSliubo struct extent_io_tree *dirty_pages, 2974acce952bSliubo int mark) 2975acce952bSliubo { 2976acce952bSliubo int ret; 2977acce952bSliubo struct page *page; 2978acce952bSliubo struct inode *btree_inode = root->fs_info->btree_inode; 2979acce952bSliubo struct extent_buffer *eb; 2980acce952bSliubo u64 start = 0; 2981acce952bSliubo u64 end; 2982acce952bSliubo u64 offset; 2983acce952bSliubo unsigned long index; 2984acce952bSliubo 2985acce952bSliubo while (1) { 2986acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 2987acce952bSliubo mark); 2988acce952bSliubo if (ret) 2989acce952bSliubo break; 2990acce952bSliubo 2991acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 2992acce952bSliubo while (start <= end) { 2993acce952bSliubo index = start >> PAGE_CACHE_SHIFT; 2994acce952bSliubo start = (u64)(index + 1) << PAGE_CACHE_SHIFT; 2995acce952bSliubo page = find_get_page(btree_inode->i_mapping, index); 2996acce952bSliubo if (!page) 2997acce952bSliubo continue; 2998acce952bSliubo offset = page_offset(page); 2999acce952bSliubo 3000acce952bSliubo spin_lock(&dirty_pages->buffer_lock); 3001acce952bSliubo eb = radix_tree_lookup( 3002acce952bSliubo &(&BTRFS_I(page->mapping->host)->io_tree)->buffer, 3003acce952bSliubo offset >> PAGE_CACHE_SHIFT); 3004acce952bSliubo spin_unlock(&dirty_pages->buffer_lock); 3005acce952bSliubo if (eb) { 3006acce952bSliubo ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY, 3007acce952bSliubo &eb->bflags); 3008acce952bSliubo atomic_set(&eb->refs, 1); 3009acce952bSliubo } 3010acce952bSliubo if (PageWriteback(page)) 3011acce952bSliubo end_page_writeback(page); 3012acce952bSliubo 3013acce952bSliubo lock_page(page); 3014acce952bSliubo if (PageDirty(page)) { 3015acce952bSliubo clear_page_dirty_for_io(page); 3016acce952bSliubo spin_lock_irq(&page->mapping->tree_lock); 3017acce952bSliubo radix_tree_tag_clear(&page->mapping->page_tree, 3018acce952bSliubo page_index(page), 3019acce952bSliubo PAGECACHE_TAG_DIRTY); 3020acce952bSliubo spin_unlock_irq(&page->mapping->tree_lock); 3021acce952bSliubo } 3022acce952bSliubo 3023acce952bSliubo page->mapping->a_ops->invalidatepage(page, 0); 3024acce952bSliubo unlock_page(page); 3025acce952bSliubo } 3026acce952bSliubo } 3027acce952bSliubo 3028acce952bSliubo return ret; 3029acce952bSliubo } 3030acce952bSliubo 3031acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 3032acce952bSliubo struct extent_io_tree *pinned_extents) 3033acce952bSliubo { 3034acce952bSliubo struct extent_io_tree *unpin; 3035acce952bSliubo u64 start; 3036acce952bSliubo u64 end; 3037acce952bSliubo int ret; 3038acce952bSliubo 3039acce952bSliubo unpin = pinned_extents; 3040acce952bSliubo while (1) { 3041acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 3042acce952bSliubo EXTENT_DIRTY); 3043acce952bSliubo if (ret) 3044acce952bSliubo break; 3045acce952bSliubo 3046acce952bSliubo /* opt_discard */ 3047acce952bSliubo ret = btrfs_error_discard_extent(root, start, end + 1 - start); 3048acce952bSliubo 3049acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 3050acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 3051acce952bSliubo cond_resched(); 3052acce952bSliubo } 3053acce952bSliubo 3054acce952bSliubo return 0; 3055acce952bSliubo } 3056acce952bSliubo 3057acce952bSliubo static int btrfs_cleanup_transaction(struct btrfs_root *root) 3058acce952bSliubo { 3059acce952bSliubo struct btrfs_transaction *t; 3060acce952bSliubo LIST_HEAD(list); 3061acce952bSliubo 3062acce952bSliubo WARN_ON(1); 3063acce952bSliubo 3064acce952bSliubo mutex_lock(&root->fs_info->trans_mutex); 3065acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 3066acce952bSliubo 3067acce952bSliubo list_splice_init(&root->fs_info->trans_list, &list); 3068acce952bSliubo while (!list_empty(&list)) { 3069acce952bSliubo t = list_entry(list.next, struct btrfs_transaction, list); 3070acce952bSliubo if (!t) 3071acce952bSliubo break; 3072acce952bSliubo 3073acce952bSliubo btrfs_destroy_ordered_operations(root); 3074acce952bSliubo 3075acce952bSliubo btrfs_destroy_ordered_extents(root); 3076acce952bSliubo 3077acce952bSliubo btrfs_destroy_delayed_refs(t, root); 3078acce952bSliubo 3079acce952bSliubo btrfs_block_rsv_release(root, 3080acce952bSliubo &root->fs_info->trans_block_rsv, 3081acce952bSliubo t->dirty_pages.dirty_bytes); 3082acce952bSliubo 3083acce952bSliubo /* FIXME: cleanup wait for commit */ 3084acce952bSliubo t->in_commit = 1; 3085acce952bSliubo t->blocked = 1; 3086acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_blocked_wait)) 3087acce952bSliubo wake_up(&root->fs_info->transaction_blocked_wait); 3088acce952bSliubo 3089acce952bSliubo t->blocked = 0; 3090acce952bSliubo if (waitqueue_active(&root->fs_info->transaction_wait)) 3091acce952bSliubo wake_up(&root->fs_info->transaction_wait); 3092acce952bSliubo mutex_unlock(&root->fs_info->trans_mutex); 3093acce952bSliubo 3094acce952bSliubo mutex_lock(&root->fs_info->trans_mutex); 3095acce952bSliubo t->commit_done = 1; 3096acce952bSliubo if (waitqueue_active(&t->commit_wait)) 3097acce952bSliubo wake_up(&t->commit_wait); 3098acce952bSliubo mutex_unlock(&root->fs_info->trans_mutex); 3099acce952bSliubo 3100acce952bSliubo mutex_lock(&root->fs_info->trans_mutex); 3101acce952bSliubo 3102acce952bSliubo btrfs_destroy_pending_snapshots(t); 3103acce952bSliubo 3104acce952bSliubo btrfs_destroy_delalloc_inodes(root); 3105acce952bSliubo 3106acce952bSliubo spin_lock(&root->fs_info->new_trans_lock); 3107acce952bSliubo root->fs_info->running_transaction = NULL; 3108acce952bSliubo spin_unlock(&root->fs_info->new_trans_lock); 3109acce952bSliubo 3110acce952bSliubo btrfs_destroy_marked_extents(root, &t->dirty_pages, 3111acce952bSliubo EXTENT_DIRTY); 3112acce952bSliubo 3113acce952bSliubo btrfs_destroy_pinned_extent(root, 3114acce952bSliubo root->fs_info->pinned_extents); 3115acce952bSliubo 3116acce952bSliubo t->use_count = 0; 3117acce952bSliubo list_del_init(&t->list); 3118acce952bSliubo memset(t, 0, sizeof(*t)); 3119acce952bSliubo kmem_cache_free(btrfs_transaction_cachep, t); 3120acce952bSliubo } 3121acce952bSliubo 3122acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 3123acce952bSliubo mutex_unlock(&root->fs_info->trans_mutex); 3124acce952bSliubo 3125acce952bSliubo return 0; 3126acce952bSliubo } 3127acce952bSliubo 3128d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops = { 31294bef0848SChris Mason .write_cache_pages_lock_hook = btree_lock_page_hook, 3130ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 31310b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 3132239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 3133239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 31340da5468fSChris Mason }; 3135